Upload 116 files

SOWA/.env.example

@@ -0,0 +1,6 @@
+# example of file for storing private and user specific environment variables, like keys or system paths
+# rename it to ".env" (excluded from version control by default)
+# .env is loaded by train.py automatically
+# hydra allows you to reference variables in .yaml configs with special syntax: ${oc.env:MY_VAR}
+
+MY_VAR="/home/user/my/system/path"

SOWA/.gitignore

@@ -0,0 +1,154 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+### VisualStudioCode
+.vscode/*
+!.vscode/settings.json
+!.vscode/tasks.json
+!.vscode/launch.json
+!.vscode/extensions.json
+*.code-workspace
+**/.vscode
+
+# JetBrains
+.idea/
+
+# Data & Models
+*.h5
+*.tar
+*.tar.gz
+
+# Lightning-Hydra-Template
+configs/local/default.yaml
+/data/
+/logs/
+.env
+
+# Aim logging
+.aim

SOWA/.pre-commit-config.yaml

@@ -0,0 +1,147 @@
+default_language_version:
+  python: python3
+
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.4.0
+    hooks:
+      # list of supported hooks: https://pre-commit.com/hooks.html
+      - id: trailing-whitespace
+      - id: end-of-file-fixer
+      - id: check-docstring-first
+      - id: check-yaml
+      - id: debug-statements
+      - id: detect-private-key
+      - id: check-executables-have-shebangs
+      - id: check-toml
+      - id: check-case-conflict
+      - id: check-added-large-files
+
+  # python code formatting
+  - repo: https://github.com/psf/black
+    rev: 23.1.0
+    hooks:
+      - id: black
+        args: [--line-length, "99"]
+
+  # python import sorting
+  - repo: https://github.com/PyCQA/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+        args: ["--profile", "black", "--filter-files"]
+
+  # python upgrading syntax to newer version
+  - repo: https://github.com/asottile/pyupgrade
+    rev: v3.3.1
+    hooks:
+      - id: pyupgrade
+        args: [--py38-plus]
+
+  # python docstring formatting
+  - repo: https://github.com/myint/docformatter
+    rev: v1.7.4
+    hooks:
+      - id: docformatter
+        args:
+          [
+            --in-place,
+            --wrap-summaries=99,
+            --wrap-descriptions=99,
+            --style=sphinx,
+            --black,
+          ]
+
+  # python docstring coverage checking
+  - repo: https://github.com/econchick/interrogate
+    rev: 1.5.0 # or master if you're bold
+    hooks:
+      - id: interrogate
+        args:
+          [
+            --verbose,
+            --fail-under=80,
+            --ignore-init-module,
+            --ignore-init-method,
+            --ignore-module,
+            --ignore-nested-functions,
+            -vv,
+          ]
+
+  # python check (PEP8), programming errors and code complexity
+  - repo: https://github.com/PyCQA/flake8
+    rev: 6.0.0
+    hooks:
+      - id: flake8
+        args:
+          [
+            "--extend-ignore",
+            "E203,E402,E501,F401,F841,RST2,RST301",
+            "--exclude",
+            "logs/*,data/*",
+          ]
+        additional_dependencies: [flake8-rst-docstrings==0.3.0]
+
+  # python security linter
+  - repo: https://github.com/PyCQA/bandit
+    rev: "1.7.5"
+    hooks:
+      - id: bandit
+        args: ["-s", "B101"]
+
+  # yaml formatting
+  - repo: https://github.com/pre-commit/mirrors-prettier
+    rev: v3.0.0-alpha.6
+    hooks:
+      - id: prettier
+        types: [yaml]
+        exclude: "environment.yaml"
+
+  # shell scripts linter
+  - repo: https://github.com/shellcheck-py/shellcheck-py
+    rev: v0.9.0.2
+    hooks:
+      - id: shellcheck
+
+  # md formatting
+  - repo: https://github.com/executablebooks/mdformat
+    rev: 0.7.16
+    hooks:
+      - id: mdformat
+        args: ["--number"]
+        additional_dependencies:
+          - mdformat-gfm
+          - mdformat-tables
+          - mdformat_frontmatter
+          # - mdformat-toc
+          # - mdformat-black
+
+  # word spelling linter
+  - repo: https://github.com/codespell-project/codespell
+    rev: v2.2.4
+    hooks:
+      - id: codespell
+        args:
+          - --skip=logs/**,data/**,*.ipynb
+          # - --ignore-words-list=abc,def
+
+  # jupyter notebook cell output clearing
+  - repo: https://github.com/kynan/nbstripout
+    rev: 0.6.1
+    hooks:
+      - id: nbstripout
+
+  # jupyter notebook linting
+  - repo: https://github.com/nbQA-dev/nbQA
+    rev: 1.6.3
+    hooks:
+      - id: nbqa-black
+        args: ["--line-length=99"]
+      - id: nbqa-isort
+        args: ["--profile=black"]
+      - id: nbqa-flake8
+        args:
+          [
+            "--extend-ignore=E203,E402,E501,F401,F841",
+            "--exclude=logs/*,data/*",
+          ]

SOWA/.project-root

@@ -0,0 +1,2 @@
+# this file is required for inferring the project root directory
+# do not delete

SOWA/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 huzongxiang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

SOWA/Makefile

@@ -0,0 +1,30 @@
+
+help:  ## Show help
+	@grep -E '^[.a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-30s\033[0m %s\n", $$1, $$2}'
+
+clean: ## Clean autogenerated files
+	rm -rf dist
+	find . -type f -name "*.DS_Store" -ls -delete
+	find . | grep -E "(__pycache__|\.pyc|\.pyo)" | xargs rm -rf
+	find . | grep -E ".pytest_cache" | xargs rm -rf
+	find . | grep -E ".ipynb_checkpoints" | xargs rm -rf
+	rm -f .coverage
+
+clean-logs: ## Clean logs
+	rm -rf logs/**
+
+format: ## Run pre-commit hooks
+	pre-commit run -a
+
+sync: ## Merge changes from main branch to your current branch
+	git pull
+	git pull origin main
+
+test: ## Run not slow tests
+	pytest -k "not slow"
+
+test-full: ## Run all tests
+	pytest
+
+train: ## Train the model
+	python src/train.py

SOWA/README.md

@@ -0,0 +1,153 @@
+<div align="center">
+
+# Soldier-Offier Window self-Attention (SOWA)
+
+<a href="https://pytorch.org/get-started/locally/"><img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-ee4c2c?logo=pytorch&logoColor=white"></a>
+<a href="https://pytorchlightning.ai/"><img alt="Lightning" src="https://img.shields.io/badge/-Lightning-792ee5?logo=pytorchlightning&logoColor=white"></a>
+<a href="https://hydra.cc/"><img alt="Config: Hydra" src="https://img.shields.io/badge/Config-Hydra-89b8cd"></a>
+<a href="https://github.com/ashleve/lightning-hydra-template"><img alt="Template" src="https://img.shields.io/badge/-Lightning--Hydra--Template-017F2F?style=flat&logo=github&labelColor=gray"></a><br>
+[![Paper](http://img.shields.io/badge/paper-arxiv.2407.03634-B31B1B.svg)](https://arxiv.org/abs/2407.03634)
+[![Conference](http://img.shields.io/badge/AnyConference-year-4b44ce.svg)](https://papers.nips.cc/paper/2020)
+
+</div>
+
+## Description
+
+<div align="center">
+  <img src="https://github.com/huzongxiang/sowa/blob/resources/fig1.png" alt="concept" style="width: 50%;">
+</div>
+
+Visual anomaly detection is critical in industrial manufacturing, but traditional methods often rely on extensive
+normal datasets and custom models, limiting scalability.
+Recent advancements in large-scale visual-language models have significantly improved zero/few-shot anomaly detection. However, these approaches may not fully utilize hierarchical features, potentially missing nuanced details. We
+introduce a window self-attention mechanism based on the
+CLIP model, combined with learnable prompts to process
+multi-level features within a Soldier-Offier Window selfAttention (SOWA) framework. Our method has been tested
+on five benchmark datasets, demonstrating superior performance by leading in 18 out of 20 metrics compared to existing state-of-the-art techniques.
+
+![architecture](https://github.com/huzongxiang/sowa/blob/resources/fig2.png)
+
+## Installation
+
+#### Pip
+
+```bash
+# clone project
+git clone https://github.com/huzongxiang/sowa
+cd sowa
+
+# [OPTIONAL] create conda environment
+conda create -n sowa python=3.9
+conda activate sowa
+
+# install pytorch according to instructions
+# https://pytorch.org/get-started/
+
+# install requirements
+pip install -r requirements.txt
+```
+
+#### Conda
+
+```bash
+# clone project
+git clone https://github.com/huzongxiang/sowa
+cd sowa
+
+# create conda environment and install dependencies
+conda env create -f environment.yaml -n sowa
+
+# activate conda environment
+conda activate sowa
+```
+
+## How to run
+
+Train model with default configuration
+
+```bash
+# train on CPU
+python src/train.py trainer=cpu data=sowa_visa model=sowa_hfwa
+
+# train on GPU
+python src/train.py trainer=gpu data=sowa_visa model=sowa_hfwa
+```
+
+## Results
+
+Comparisons with few-shot (K=4) anomaly detection methods on datasets of MVTec-AD, Visa, BTAD, DAGM and DTD Synthetic. 
+| Metric    | Dataset        | WinCLIP     | April-GAN        | Ours        |
+|-----------|----------------|-------------|-------------|-------------|
+| AC AUROC  | MVTec-AD       | 95.2±1.3    | 92.8±0.2    | 96.8±0.3    |
+|           | Visa           | 87.3±1.8    | 92.6±0.4    | 92.9±0.2    |
+|           | BTAD           | 87.0±0.2    | 92.1±0.2    | 94.8±0.2    |
+|           | DAGM           | 93.8±0.2    | 96.2±1.1    | 98.9±0.3    |
+|           | DTD-Synthetic  | 98.1±0.2    | 98.5±0.1    | 99.1±0.0    |
+| AC AP     | MVTec-AD       | 97.3±0.6    | 96.3±0.1    | 98.3±0.3    |
+|           | Visa           | 88.8±1.8    | 94.5±0.3    | 94.5±0.2    |
+|           | BTAD           | 86.8±0.0    | 95.2±0.5    | 95.5±0.7    |
+|           | DAGM           | 83.8±1.1    | 86.7±4.5    | 95.2±1.7    |
+|           | DTD-Synthetic  | 99.1±0.1    | 99.4±0.0    | 99.6±0.0    |
+| AS AUROC  | MVTec-AD       | 96.2±0.3    | 95.9±0.0    | 95.7±0.1    |
+|           | Visa           | 97.2±0.2    | 96.2±0.0    | 97.1±0.0    |
+|           | BTAD           | 95.8±0.0    | 94.4±0.1    | 97.1±0.0    |
+|           | DAGM           | 93.8±0.1    | 88.9±0.4    | 96.9±0.0    |
+|           | DTD-Synthetic  | 96.8±0.2    | 96.7±0.0    | 98.7±0.0    |
+| AS AUPRO  | MVTec-AD       | 89.0±0.8    | 91.8±0.1    | 92.4±0.2    |
+|           | Visa           | 87.6±0.9    | 90.2±0.1    | 91.4±0.0    |
+|           | BTAD           | 66.6±0.2    | 78.2±0.1    | 81.2±0.2    |
+|           | DAGM           | 82.4±0.3    | 77.8±0.9    | 94.4±0.1    |
+|           | DTD-Synthetic  | 90.1±0.5    | 92.2±0.0    | 96.6±0.1    | 
+
+​<!-- 零宽空格 -->
+
+Performance Comparison on MVTec-AD and Visa Datasets. 
+| Method        | Source                  | MVTec-AD AC AUROC | MVTec-AD AS AUROC | MVTec-AD AS PRO | Visa AC AUROC | Visa AS AUROC | Visa AS PRO |
+|---------------|-------------------------|-------------------|-------------------|-----------------|---------------|---------------|-------------|
+| SPADE         | arXiv 2020              | 84.8±2.5          | 92.7±0.3          | 87.0±0.5        | 81.7±3.4      | 96.6±0.3      | 87.3±0.8    |
+| PaDiM         | ICPR 2021               | 80.4±2.4          | 92.6±0.7          | 81.3±1.9        | 72.8±2.9      | 93.2±0.5      | 72.6±1.9    |
+| PatchCore     | CVPR 2022               | 88.8±2.6          | 94.3±0.5          | 84.3±1.6        | 85.3±2.1      | 96.8±0.3      | 84.9±1.4    |
+| WinCLIP       | CVPR 2023               | 95.2±1.3          | 96.2±0.3          | 89.0±0.8        | 87.3±1.8      | 97.2±0.2      | 87.6±0.9    |
+| April-GAN     | CVPR 2023 VAND workshop | 92.8±0.2          | 95.9±0.0          | 91.8±0.1        | 92.6±0.4      | 96.2±0.0      | 90.2±0.1    |
+| PromptAD      | CVPR 2024               | 96.6±0.9          | 96.5±0.2          | -               | 89.1±1.7      | 97.4±0.3      | -           |
+| InCTRL        | CVPR 2024               | 94.5±1.8          | -                 | -               | 87.7±1.9      | -             | -           |
+| SOWA          | Ours                    | 96.8±0.3          | 95.7±0.1          | 92.4±0.2        | 92.9±0.2      | 97.1±0.0      | 91.4±0.0    | 
+
+
+​<!-- 零宽空格 -->
+
+Comparisons with few-shot anomaly detection methods on datasets of MVTec-AD, Visa, BTAD, DAGM and DTD Synthetic. 
+<div align="center">
+  <img src="https://github.com/huzongxiang/sowa/blob/resources/fig5.png" alt="few-shot" style="width: 70%;">
+</div>
+
+
+## Visualization
+Visualization results under the few-shot setting (K=4). 
+<div align="center">
+  <img src="https://github.com/huzongxiang/sowa/blob/resources/fig6.png" alt="concept" style="width: 70%;">
+</div>
+
+
+## Mechanism
+Hierarchical Results on MVTec-AD Dataset. A set of images showing the real outputs of the model, illustrating how different layers (H1 to H4) process various feature modes. Each row represents a different sample, with columns showing the original image, segmentation mask, heatmap, and feature outputs from H1 to H4, and fusion.
+![mechanism](https://github.com/huzongxiang/sowa/blob/resources/fig7.png)
+
+
+## Inference Speed
+Inference performance comparison of different methods on a single NVIDIA RTX3070 8GB GPU.
+<div align="center">
+  <img src="https://github.com/huzongxiang/sowa/blob/resources/fig9.png" alt="speed" style="width: 80%;">
+</div>
+
+
+## Citation
+Please cite the following paper if this work helps your project: 
+```
+@article{hu2024sowa,
+  title={SOWA: Adapting Hierarchical Frozen Window Self-Attention to Visual-Language Models for Better Anomaly Detection},
+  author={Hu, Zongxiang and Zhang, zhaosheng},
+  journal={arXiv preprint arXiv:2407.03634},
+  year={2024}
+}
+```

SOWA/configs/__init__.py

@@ -0,0 +1 @@
+# this file is needed here to include configs when building project as a package

SOWA/configs/callbacks/default.yaml

@@ -0,0 +1,27 @@
+defaults:
+  - model_checkpoint
+  - early_stopping
+  - model_summary
+  - rich_progress_bar
+  - visualization
+  - _self_
+
+model_checkpoint:
+  dirpath: ${paths.output_dir}/checkpoints
+  filename: "epoch_{epoch:03d}"
+  monitor: "train/loss"
+  mode: "min"
+  save_last: True
+  auto_insert_metric_name: False
+
+early_stopping:
+  monitor: "train/loss"
+  patience: 10
+  mode: "min"
+
+model_summary:
+  max_depth: -1
+
+visualization:
+  dirpath: ${paths.output_dir}/visualizations
+  visualize: True

SOWA/configs/callbacks/early_stopping.yaml

@@ -0,0 +1,15 @@
+# https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.EarlyStopping.html
+
+early_stopping:
+  _target_: lightning.pytorch.callbacks.EarlyStopping
+  monitor: ??? # quantity to be monitored, must be specified !!!
+  min_delta: 0. # minimum change in the monitored quantity to qualify as an improvement
+  patience: 3 # number of checks with no improvement after which training will be stopped
+  verbose: False # verbosity mode
+  mode: "min" # "max" means higher metric value is better, can be also "min"
+  strict: True # whether to crash the training if monitor is not found in the validation metrics
+  check_finite: True # when set True, stops training when the monitor becomes NaN or infinite
+  stopping_threshold: null # stop training immediately once the monitored quantity reaches this threshold
+  divergence_threshold: null # stop training as soon as the monitored quantity becomes worse than this threshold
+  check_on_train_epoch_end: null # whether to run early stopping at the end of the training epoch
+  # log_rank_zero_only: False  # this keyword argument isn't available in stable version

SOWA/configs/callbacks/model_checkpoint.yaml

@@ -0,0 +1,17 @@
+# https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.ModelCheckpoint.html
+
+model_checkpoint:
+  _target_: lightning.pytorch.callbacks.ModelCheckpoint
+  dirpath: null # directory to save the model file
+  filename: null # checkpoint filename
+  monitor: null # name of the logged metric which determines when model is improving
+  verbose: False # verbosity mode
+  save_last: null # additionally always save an exact copy of the last checkpoint to a file last.ckpt
+  save_top_k: 1 # save k best models (determined by above metric)
+  mode: "min" # "max" means higher metric value is better, can be also "min"
+  auto_insert_metric_name: True # when True, the checkpoints filenames will contain the metric name
+  save_weights_only: False # if True, then only the model’s weights will be saved
+  every_n_train_steps: null # number of training steps between checkpoints
+  train_time_interval: null # checkpoints are monitored at the specified time interval
+  every_n_epochs: null # number of epochs between checkpoints
+  save_on_train_epoch_end: null # whether to run checkpointing at the end of the training epoch or the end of validation

SOWA/configs/callbacks/model_summary.yaml

@@ -0,0 +1,5 @@
+# https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.RichModelSummary.html
+
+model_summary:
+  _target_: lightning.pytorch.callbacks.RichModelSummary
+  max_depth: 1 # the maximum depth of layer nesting that the summary will include

SOWA/configs/callbacks/rich_progress_bar.yaml

@@ -0,0 +1,4 @@
+# https://lightning.ai/docs/pytorch/latest/api/lightning.pytorch.callbacks.RichProgressBar.html
+
+rich_progress_bar:
+  _target_: lightning.pytorch.callbacks.RichProgressBar

SOWA/configs/callbacks/visualization.yaml

@@ -0,0 +1,4 @@
+visualization:
+  _target_: src.models.components.callback.AnomalyVisualizationCallback
+  dirpath: ${paths.output_dir}/visualizations
+  visualize: True

SOWA/configs/data/sowa_infer.yaml

@@ -0,0 +1,56 @@
+_target_: src.data.anomaly_clip_datamodule.AnomalyCLIPDataModule
+data_dir:
+  train: /home/hzx/Projects/Data/MVTec-AD
+  valid: /home/hzx/Projects/Data/MVTec-AD
+  # test: /home/hzx/Projects/Data/BTAD
+  # test: /home/hzx/Projects/Data/DAGM
+  test: /home/hzx/Projects/Data/DTD-Synthetic
+  # test: /home/hzx/Projects/Data/MPDD
+  # test: /home/hzx/Projects/Data/SDD
+dataset:
+  train:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+    aug_rate: 0.2
+  valid:
+    _target_: src.data.components.anomal_dataset.VisaDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  test:
+    _target_: src.data.components.anomal_dataset.VisaDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  kshot:
+    _target_: src.data.components.kshot_dataset.VisaKShotDataset
+    _partial_: true
+    k_shot: 4
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+image_size: 336
+num_workers: 6
+pin_memory: False
+batch_size: 8

SOWA/configs/data/sowa_mvt.yaml

@@ -0,0 +1,53 @@
+_target_: src.data.anomaly_clip_datamodule.AnomalyCLIPDataModule
+data_dir:
+  train: /home/hzx/Projects/Data/Visa
+  valid: /home/hzx/Projects/Data/MVTec-AD
+  test: /home/hzx/Projects/Data/MVTec-AD
+dataset:
+  train:
+    _target_: src.data.components.anomal_dataset.VisaDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  valid:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+    aug_rate: 0.0
+  test:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+    aug_rate: 0.0
+  kshot:
+    _target_: src.data.components.kshot_dataset.MVTecKShotDataset
+    _partial_: true
+    k_shot: 1
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+image_size: 336
+num_workers: 6
+pin_memory: False
+batch_size: 8

SOWA/configs/data/sowa_overfit.yaml

@@ -0,0 +1,51 @@
+_target_: src.data.anomaly_clip_datamodule.AnomalyCLIPDataModule
+data_dir:
+  train: /home/hzx/Projects/Data/MVTec-AD
+  valid: /home/hzx/Projects/Data/MVTec-AD
+  test: /home/hzx/Projects/Data/MVTec-AD
+dataset:
+  train:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  valid:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  test:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  kshot:
+    _target_: src.data.components.kshot_dataset.MVTecKShotDataset
+    _partial_: true
+    k_shot: 1
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+image_size: 336
+num_workers: 6
+pin_memory: False
+batch_size: 8

SOWA/configs/data/sowa_visa.yaml

@@ -0,0 +1,52 @@
+_target_: src.data.anomaly_clip_datamodule.AnomalyCLIPDataModule
+data_dir:
+  train: /home/hzx/Projects/Data/MVTec-AD
+  valid: /home/hzx/Projects/Data/Visa
+  test: /home/hzx/Projects/Data/Visa
+dataset:
+  train:
+    _target_: src.data.components.anomal_dataset.MVTecDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+    aug_rate: 0.2
+  valid:
+    _target_: src.data.components.anomal_dataset.VisaDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  test:
+    _target_: src.data.components.anomal_dataset.VisaDataset
+    _partial_: true
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+  kshot:
+    _target_: src.data.components.kshot_dataset.VisaKShotDataset
+    _partial_: true
+    k_shot: 1
+    transform:
+      _target_: src.data.components.transform.ImageTransform
+      image_size: 336
+    mask_transform:
+      _target_: src.data.components.transform.MaskTransform
+      image_size: ${data.image_size}
+    preload: false
+image_size: 336
+num_workers: 6
+pin_memory: False
+batch_size: 8

SOWA/configs/debug/default.yaml

@@ -0,0 +1,35 @@
+# @package _global_
+
+# default debugging setup, runs 1 full epoch
+# other debugging configs can inherit from this one
+
+# overwrite task name so debugging logs are stored in separate folder
+task_name: "debug"
+
+# disable callbacks and loggers during debugging
+callbacks: null
+logger: null
+
+extras:
+  ignore_warnings: False
+  enforce_tags: False
+
+# sets level of all command line loggers to 'DEBUG'
+# https://hydra.cc/docs/tutorials/basic/running_your_app/logging/
+hydra:
+  job_logging:
+    root:
+      level: DEBUG
+
+  # use this to also set hydra loggers to 'DEBUG'
+  # verbose: True
+
+trainer:
+  max_epochs: 1
+  accelerator: cpu # debuggers don't like gpus
+  devices: 1 # debuggers don't like multiprocessing
+  detect_anomaly: true # raise exception if NaN or +/-inf is detected in any tensor
+
+data:
+  num_workers: 0 # debuggers don't like multiprocessing
+  pin_memory: False # disable gpu memory pin

SOWA/configs/debug/fdr.yaml

@@ -0,0 +1,9 @@
+# @package _global_
+
+# runs 1 train, 1 validation and 1 test step
+
+defaults:
+  - default
+
+trainer:
+  fast_dev_run: true

SOWA/configs/debug/limit.yaml

@@ -0,0 +1,12 @@
+# @package _global_
+
+# uses only 1% of the training data and 5% of validation/test data
+
+defaults:
+  - default
+
+trainer:
+  max_epochs: 3
+  limit_train_batches: 0.01
+  limit_val_batches: 0.05
+  limit_test_batches: 0.05

SOWA/configs/debug/overfit.yaml

@@ -0,0 +1,13 @@
+# @package _global_
+
+# overfits to 3 batches
+
+defaults:
+  - default
+
+trainer:
+  max_epochs: 20
+  overfit_batches: 3
+
+# model ckpt and early stopping need to be disabled during overfitting
+callbacks: null

SOWA/configs/debug/profiler.yaml

@@ -0,0 +1,12 @@
+# @package _global_
+
+# runs with execution time profiling
+
+defaults:
+  - default
+
+trainer:
+  max_epochs: 1
+  profiler: "simple"
+  # profiler: "advanced"
+  # profiler: "pytorch"

SOWA/configs/eval.yaml

@@ -0,0 +1,27 @@
+# @package _global_
+
+defaults:
+  - _self_
+  - data: anomaly_clip # choose datamodule with `test_dataloader()` for evaluation
+  - model: anomaly_clip
+  - callbacks: default
+  - logger: many_loggers
+  - trainer: default
+  - paths: default
+  - extras: default
+  - hydra: default
+
+  # information of object and prompt template
+  - prompt: default
+
+task_name: "eval"
+
+tags: ["dev"]
+
+# seed for random number generators in pytorch, numpy and python.random
+seed: 42
+
+# passing checkpoint path is necessary for evaluation
+# ckpt_path: /home/hzx/Projects/SPARC/logs/train/runs/2024-06-14_16-32-14/checkpoints/epoch_000.ckpt
+# ckpt_path: /home/hzx/Projects/Weight/mvtech-kshot/learnable/0-shot/2024-05-27_13-08-55/checkpoints/epoch_000.ckpt
+ckpt_path: /home/hzx/Projects/Weight/visa-kshot/learnable/0-shot/2024-05-27_11-07-16/checkpoints/epoch_000.ckpt

SOWA/configs/experiment/example.yaml

@@ -0,0 +1,194 @@
+task_name: train
+tags:
+- dev
+train: true
+test: true
+ckpt_path: null
+seed: 2025
+data:
+  _target_: src.data.anomaly_clip_datamodule.AnomalyCLIPDataModule
+  data_dir:
+    train: /home/hzx/Projects/Data/Visa
+    valid: /home/hzx/Projects/Data/MVTec-AD
+    test: /home/hzx/Projects/Data/MVTec-AD
+  dataset:
+    train:
+      _target_: src.data.components.anomal_dataset.VisaDataset
+      _partial_: true
+      transform:
+        _target_: src.data.components.transform.ImageTransform
+        image_size: 336
+      mask_transform:
+        _target_: src.data.components.transform.MaskTransform
+        image_size: ${data.image_size}
+      preload: false
+    valid:
+      _target_: src.data.components.anomal_dataset.MVTecDataset
+      _partial_: true
+      transform:
+        _target_: src.data.components.transform.ImageTransform
+        image_size: 336
+      mask_transform:
+        _target_: src.data.components.transform.MaskTransform
+        image_size: ${data.image_size}
+      preload: false
+      aug_rate: 0.0
+    test:
+      _target_: src.data.components.anomal_dataset.MVTecDataset
+      _partial_: true
+      transform:
+        _target_: src.data.components.transform.ImageTransform
+        image_size: 336
+      mask_transform:
+        _target_: src.data.components.transform.MaskTransform
+        image_size: ${data.image_size}
+      preload: false
+      aug_rate: 0.0
+    kshot:
+      _target_: src.data.components.kshot_dataset.MVTecKShotDataset
+      _partial_: true
+      k_shot: 1
+      transform:
+        _target_: src.data.components.transform.ImageTransform
+        image_size: 336
+      mask_transform:
+        _target_: src.data.components.transform.MaskTransform
+        image_size: ${data.image_size}
+      preload: false
+  image_size: 336
+  num_workers: 4
+  pin_memory: false
+  batch_size: 8
+model:
+  _target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+  optimizer:
+    _target_: torch.optim.AdamW
+    _partial_: true
+    lr: 0.001
+    weight_decay: 0.2
+  scheduler:
+    _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+    _partial_: true
+    mode: min
+    factor: 0.1
+    patience: 5
+  net:
+    _target_: src.models.components.anomaly_clip.AnomalyCLIP
+    arch: ViT-L/14@336px
+    image_size: 336
+    class_names:
+    - object
+    temperature: 0.05
+    prompt_length: 24
+    context_length: 77
+    truncate: false
+    feature_map_idx:
+    - 5
+    - 11
+    - 17
+    - 23
+    share_weight: false
+    state_template:
+      normal:
+      - '{}'
+      anomaly:
+      - damaged {}
+    tokenizer:
+      _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+    adapter:
+      _target_: src.models.components.adapter.BasicLayer
+      _partial_: true
+      input_resolution:
+      - 24
+      - 24
+      window_size: 6
+      depth: 1
+      num_heads: 8
+      hidden_features: null
+      cpb_dim: 64
+      value_only: true
+      drop: 0.0
+      attn_drop: 0.2
+  loss:
+    cross_entropy:
+      _target_: torch.nn.CrossEntropyLoss
+    focal:
+      _target_: src.models.components.loss.FocalLoss
+    dice:
+      _target_: src.models.components.loss.BinaryDiceLoss
+  k_shot: false
+  enable_validation: false
+  compile: false
+callbacks:
+  model_checkpoint:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    dirpath: ${paths.output_dir}/checkpoints
+    filename: epoch_{epoch:03d}
+    monitor: train/loss
+    verbose: false
+    save_last: true
+    save_top_k: 1
+    mode: min
+    auto_insert_metric_name: false
+    save_weights_only: false
+    every_n_train_steps: null
+    train_time_interval: null
+    every_n_epochs: null
+    save_on_train_epoch_end: null
+  early_stopping:
+    _target_: lightning.pytorch.callbacks.EarlyStopping
+    monitor: train/loss
+    min_delta: 0.0
+    patience: 10
+    verbose: false
+    mode: min
+    strict: true
+    check_finite: true
+    stopping_threshold: null
+    divergence_threshold: null
+    check_on_train_epoch_end: null
+  model_summary:
+    _target_: lightning.pytorch.callbacks.RichModelSummary
+    max_depth: -1
+  rich_progress_bar:
+    _target_: lightning.pytorch.callbacks.RichProgressBar
+  visualization:
+    _target_: src.models.components.callback.AnomalyVisualizationCallback
+    dirpath: ${paths.output_dir}/visualizations
+    visualize: true
+  visulization:
+    dirpath: ${paths.output_dir}/visualizations
+    visualize: true
+logger:
+  wandb:
+    _target_: lightning.pytorch.loggers.wandb.WandbLogger
+    save_dir: ${paths.output_dir}
+    offline: false
+    id: null
+    anonymous: null
+    project: mvt_optuna
+    log_model: false
+    prefix: ''
+    group: ''
+    tags: []
+    job_type: ''
+trainer:
+  _target_: lightning.pytorch.trainer.Trainer
+  default_root_dir: ${paths.output_dir}
+  min_epochs: 1
+  max_epochs: 2
+  accelerator: gpu
+  devices: 1
+  check_val_every_n_epoch: 1
+  deterministic: false
+paths:
+  root_dir: ${oc.env:PROJECT_ROOT}
+  data_dir: ${paths.root_dir}/data/
+  log_dir: ${paths.root_dir}/logs/
+  output_dir: ${hydra:runtime.output_dir}
+  work_dir: ${hydra:runtime.cwd}
+extras:
+  ignore_warnings: false
+  enforce_tags: true
+  print_config: true
+optimized_metric: test/objective

SOWA/configs/extras/default.yaml

@@ -0,0 +1,8 @@
+# disable python warnings if they annoy you
+ignore_warnings: False
+
+# ask user for tags if none are provided in the config
+enforce_tags: True
+
+# pretty print config tree at the start of the run using Rich library
+print_config: True

SOWA/configs/hparams_search/anomaly_clip_optuna.yaml

@@ -0,0 +1,61 @@
+# @package _global_
+
+# example hyperparameter optimization of some experiment with Optuna:
+# python train.py -m hparams_search=mnist_optuna experiment=example
+
+defaults:
+  - override /hydra/sweeper: optuna
+
+# choose metric which will be optimized by Optuna
+# make sure this is the correct name of some metric logged in lightning module!
+optimized_metric: test/objective
+
+# here we define Optuna hyperparameter search
+# it optimizes for value returned from function with @hydra.main decorator
+# docs: https://hydra.cc/docs/next/plugins/optuna_sweeper
+hydra:
+  mode: "MULTIRUN" # set hydra to multirun by default if this config is attached
+
+  sweeper:
+    _target_: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper
+
+    # storage URL to persist optimization results
+    # for example, you can use SQLite if you set 'sqlite:///example.db'
+    storage: null
+
+    # name of the study to persist optimization results
+    study_name: null
+
+    # number of parallel workers
+    n_jobs: 1
+
+    # 'minimize' or 'maximize' the objective
+    direction: maximize
+
+    # total number of runs that will be executed
+    n_trials: 50
+
+    # choose Optuna hyperparameter sampler
+    # you can choose bayesian sampler (tpe), random search (without optimization), grid sampler, and others
+    # docs: https://optuna.readthedocs.io/en/stable/reference/samplers.html
+    sampler:
+      _target_: optuna.samplers.TPESampler
+      seed: 1234
+      n_startup_trials: 50 # number of random sampling runs before optimization starts
+
+    # define hyperparameter search space
+    params:
+      trainer.max_epochs: choice(1, 5)
+      # model.optimizer.lr: choice(0.0001, 0.001)
+      # model.net.temperature: choice(0.1, 0.05)
+      model.net.prompt_length: choice(8, 12, 16, 24)
+      model.net.share_weight: choice(true, false)
+      model.net.feature_map_idx : choice([5, 11, 17, 23], [0, 11, 23])
+      # model.net.adapter.hidden_features: choice([1024])
+      model.net.adapter.window_size: choice(6, 12, 24)
+      model.net.adapter.depth: choice(1, 2)
+      model.net.adapter.num_heads: choice(8)
+      # model.net.adapter.cpb_dim: choice(64, 128, 512)
+      model.net.adapter.value_only: choice(true, false)
+      model.net.adapter.drop: choice(0.0, 0.1, 0.2)
+      model.net.adapter.attn_drop: choice(0.0, 0.1, 0.2)

SOWA/configs/hydra/default.yaml

@@ -0,0 +1,19 @@
+# https://hydra.cc/docs/configure_hydra/intro/
+
+# enable color logging
+defaults:
+  - override hydra_logging: default
+  - override job_logging: default
+
+# output directory, generated dynamically on each run
+run:
+  dir: ${paths.log_dir}/${task_name}/runs/${now:%Y-%m-%d}_${now:%H-%M-%S}
+sweep:
+  dir: ${paths.log_dir}/${task_name}/multiruns/${now:%Y-%m-%d}_${now:%H-%M-%S}
+  subdir: ${hydra.job.num}
+
+job_logging:
+  handlers:
+    file:
+      # Incorporates fix from https://github.com/facebookresearch/hydra/pull/2242
+      filename: ${hydra.runtime.output_dir}/${task_name}.log

SOWA/configs/logger/aim.yaml

@@ -0,0 +1,28 @@
+# https://aimstack.io/
+
+# example usage in lightning module:
+# https://github.com/aimhubio/aim/blob/main/examples/pytorch_lightning_track.py
+
+# open the Aim UI with the following command (run in the folder containing the `.aim` folder):
+# `aim up`
+
+aim:
+  _target_: aim.pytorch_lightning.AimLogger
+  repo: ${paths.root_dir} # .aim folder will be created here
+  # repo: "aim://ip_address:port" # can instead provide IP address pointing to Aim remote tracking server which manages the repo, see https://aimstack.readthedocs.io/en/latest/using/remote_tracking.html#
+
+  # aim allows to group runs under experiment name
+  experiment: null # any string, set to "default" if not specified
+
+  train_metric_prefix: "train/"
+  val_metric_prefix: "val/"
+  test_metric_prefix: "test/"
+
+  # sets the tracking interval in seconds for system usage metrics (CPU, GPU, memory, etc.)
+  system_tracking_interval: 10 # set to null to disable system metrics tracking
+
+  # enable/disable logging of system params such as installed packages, git info, env vars, etc.
+  log_system_params: true
+
+  # enable/disable tracking console logs (default value is true)
+  capture_terminal_logs: false # set to false to avoid infinite console log loop issue https://github.com/aimhubio/aim/issues/2550

SOWA/configs/logger/comet.yaml

@@ -0,0 +1,12 @@
+# https://www.comet.ml
+
+comet:
+  _target_: lightning.pytorch.loggers.comet.CometLogger
+  api_key: ${oc.env:COMET_API_TOKEN} # api key is loaded from environment variable
+  save_dir: "${paths.output_dir}"
+  project_name: "lightning-hydra-template"
+  rest_api_key: null
+  # experiment_name: ""
+  experiment_key: null # set to resume experiment
+  offline: False
+  prefix: ""

SOWA/configs/logger/csv.yaml

@@ -0,0 +1,7 @@
+# csv logger built in lightning
+
+csv:
+  _target_: lightning.pytorch.loggers.csv_logs.CSVLogger
+  save_dir: "${paths.output_dir}"
+  name: "csv/"
+  prefix: ""

SOWA/configs/logger/many_loggers.yaml

@@ -0,0 +1,9 @@
+# train with many loggers at once
+
+defaults:
+  # - comet
+  - csv
+  # - mlflow
+  # - neptune
+  # - tensorboard
+  - wandb

SOWA/configs/logger/mlflow.yaml

@@ -0,0 +1,12 @@
+# https://mlflow.org
+
+mlflow:
+  _target_: lightning.pytorch.loggers.mlflow.MLFlowLogger
+  # experiment_name: ""
+  # run_name: ""
+  tracking_uri: ${paths.log_dir}/mlflow/mlruns # run `mlflow ui` command inside the `logs/mlflow/` dir to open the UI
+  tags: null
+  # save_dir: "./mlruns"
+  prefix: ""
+  artifact_location: null
+  # run_id: ""

SOWA/configs/logger/neptune.yaml

@@ -0,0 +1,9 @@
+# https://neptune.ai
+
+neptune:
+  _target_: lightning.pytorch.loggers.neptune.NeptuneLogger
+  api_key: ${oc.env:NEPTUNE_API_TOKEN} # api key is loaded from environment variable
+  project: username/lightning-hydra-template
+  # name: ""
+  log_model_checkpoints: True
+  prefix: ""

SOWA/configs/logger/tensorboard.yaml

@@ -0,0 +1,10 @@
+# https://www.tensorflow.org/tensorboard/
+
+tensorboard:
+  _target_: lightning.pytorch.loggers.tensorboard.TensorBoardLogger
+  save_dir: "${paths.output_dir}/tensorboard/"
+  name: null
+  log_graph: False
+  default_hp_metric: True
+  prefix: ""
+  # version: ""

SOWA/configs/logger/wandb.yaml

@@ -0,0 +1,16 @@
+# https://wandb.ai
+
+wandb:
+  _target_: lightning.pytorch.loggers.wandb.WandbLogger
+  # name: "" # name of the run (normally generated by wandb)
+  save_dir: "${paths.output_dir}"
+  offline: False
+  id: null # pass correct id to resume experiment!
+  anonymous: null # enable anonymous logging
+  project: mvt_optuna
+  log_model: False # upload lightning ckpts
+  prefix: "" # a string to put at the beginning of metric keys
+  # entity: "" # set to name of your wandb team
+  group: ""
+  tags: []
+  job_type: ""

SOWA/configs/model/sowa_hfwa.yaml

@@ -0,0 +1,71 @@
+_target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+
+optimizer:
+  _target_: torch.optim.AdamW
+  _partial_: true
+  lr: 0.001
+  weight_decay: 0.2
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  _partial_: true
+  mode: min
+  factor: 0.1
+  patience: 5
+
+# scheduler:
+#   _target_: src.models.components.scheduler.WarmupCosineAnnealingLR
+#   _partial_: true
+#   warmup_epochs: 10
+#   total_epoch: 50
+
+net:
+  _target_: src.models.components.anomaly_clip.AnomalyCLIP
+  arch: ViT-L/14@336px
+  image_size: 336
+  class_names: ["object"]
+  # class_names: ${prompt.class_names}
+  temperature: 0.07                     # softmax
+  prompt_length: 12                     # length of learnable prompts
+  context_length: 77                    # defaut 77 for openai clip
+  truncate: false
+  feature_map_idx: [5, 11, 17, 23]      # [0, 12, 23] [6, 12, 18] [5, 11, 17, 23] index of resnetblock in ViT
+  share_weight: true                    # whether the adapter shares weights for different feature maps
+  # state_template: ${prompt.state_template}
+  state_template: 
+    normal: ["{}"]
+    anomaly: ["damaged {}"]
+  tokenizer:
+    _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+  adapter:
+    _target_: src.models.components.adapter.BasicLayer
+    _partial_: true
+    input_resolution: [24, 24]          # (image_size - kerner_size) / stride + 1. eg. 24 = (224 - 14) / 14 + 1
+    window_size: 12
+    depth: 1                            # if depth < 2, thers is no window shift
+    num_heads: 8
+    hidden_features: null               # set null, same as nn.Linear
+    cpb_dim: 64
+    value_only: true
+    drop: 0.0
+    attn_drop: 0.1
+    # shift_size: 1
+  fusion:
+    _target_: src.models.components.cross_modal.DotProductFusion
+  embedding_dim: 768                    # clip fusion featrue dim, default 768, only effective for non null
+
+loss:
+  cross_entropy:
+    _target_: torch.nn.CrossEntropyLoss
+  focal:
+    _target_: src.models.components.loss.FocalLoss
+  dice:
+    _target_: src.models.components.loss.BinaryDiceLoss
+
+k_shot: false
+
+filter: true
+
+enable_validation: false
+
+compile: false

SOWA/configs/model/sowa_linear.yaml

@@ -0,0 +1,63 @@
+_target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+
+optimizer:
+  _target_: torch.optim.AdamW
+  _partial_: true
+  lr: 0.001
+  weight_decay: 0.2
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  _partial_: true
+  mode: min
+  factor: 0.1
+  patience: 5
+
+net:
+  _target_: src.models.components.anomaly_clip.AnomalyCLIP
+  arch: ViT-L/14@336px
+  image_size: 336
+  class_names: ["object"]
+  # class_names: ${prompt.class_names}
+  temperature: 0.07                     # softmax
+  prompt_length: 12                     # length of learnable prompts
+  context_length: 77                    # defaut 77 for openai clip
+  truncate: false
+  feature_map_idx: [5, 11, 17, 23]      # [0, 12, 23] [6, 12, 18] [5, 11, 17, 23] index of resnetblock in ViT
+  share_weight: false                   # whether the adapter shares weights for different feature maps
+  # state_template: ${prompt.state_template}
+  state_template: 
+    normal: ["{}"]
+    anomaly: ["damaged {}"]
+  tokenizer:
+    _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+  adapter:
+    # _target_: torch.nn.Linear
+    # in_features: 1024                   # clip vit feature dim, defaut 1024 for openai clip
+    # out_features: 1024
+    # bias: false
+    _target_: src.models.components.adapter.Linear
+    in_features: 1024                     # clip vit feature dim, defaut 1024 for openai clip
+    out_features: 1024
+    hidden_features: null                 # set null, same as nn.Linear
+    dropout_prob: 0.0
+    bias: false
+  fusion:
+    _target_: src.models.components.cross_modal.DotProductFusion
+  embedding_dim: null                     # clip fusion featrue dim, only effective for learnable
+
+loss:
+  cross_entropy:
+    _target_: torch.nn.CrossEntropyLoss
+  focal:
+    _target_: src.models.components.loss.FocalLoss
+  dice:
+    _target_: src.models.components.loss.BinaryDiceLoss
+
+k_shot: false
+
+filter: true
+
+enable_validation: false
+
+compile: false

SOWA/configs/model/sparc_hfwa.yaml

@@ -0,0 +1,75 @@
+_target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+
+optimizer:
+  _target_: torch.optim.AdamW
+  _partial_: true
+  lr: 0.001
+  weight_decay: 0.2
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  _partial_: true
+  mode: min
+  factor: 0.1
+  patience: 5
+
+# scheduler:
+#   _target_: src.models.components.scheduler.WarmupCosineAnnealingLR
+#   _partial_: true
+#   warmup_epochs: 10
+#   total_epoch: 50
+
+net:
+  _target_: src.models.components.sparc.SPARC
+  arch: ViT-L/14@336px
+  image_size: 336
+  temperature: 0.07                     # softmax
+  feature_map_idx: [5, 11, 17, 23]      # [0, 12, 23] [6, 12, 18] [5, 11, 17, 23] index of resnetblock in ViT
+  prompt_learner:
+    _target_: src.models.components.coop.AnomalyPromptLearner
+    _partial_: true
+    tokenizer:
+      _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+    prompt_length: 12                     # length of learnable prompts
+    context_length: 77                    # defaut 77 for openai clip
+    truncate: false
+    class_names: ["object"]
+    # class_names: ${prompt.class_names}
+    # state_template: ${prompt.state_template}
+    state_template: 
+      normal: ["{}"]
+      anomaly: ["damaged {}"]
+  text_encoder:
+    _target_: src.models.components.text_encoder.TextMapEncoder
+    _partial_: true
+  adapter:
+    _target_: src.models.components.adapter.BasicLayer
+    _partial_: true
+    input_resolution: [24, 24]          # (image_size - kerner_size) / stride + 1. eg. 24 = (224 - 14) / 14 + 1
+    window_size: 12
+    depth: 1                            # if depth < 2, thers is no window shift
+    num_heads: 8
+    hidden_features: null               # set null, same as nn.Linear
+    cpb_dim: 64
+    value_only: true
+    drop: 0.0
+    attn_drop: 0.1
+  fusion:
+    _target_: src.models.components.cross_modal.DotProductFusion
+  embedding_dim: 768                    # clip fusion featrue dim, default 768, only effective for non null
+
+loss:
+  cross_entropy:
+    _target_: torch.nn.CrossEntropyLoss
+  focal:
+    _target_: src.models.components.loss.FocalLoss
+  dice:
+    _target_: src.models.components.loss.BinaryDiceLoss
+
+k_shot: false
+
+filter: true
+
+enable_validation: false
+
+compile: false

SOWA/configs/model/sparc_linear.yaml

@@ -0,0 +1,74 @@
+_target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+
+optimizer:
+  _target_: torch.optim.AdamW
+  _partial_: true
+  lr: 0.001
+  weight_decay: 0.2
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  _partial_: true
+  mode: min
+  factor: 0.1
+  patience: 5
+
+# scheduler:
+#   _target_: src.models.components.scheduler.WarmupCosineAnnealingLR
+#   _partial_: true
+#   warmup_epochs: 10
+#   total_epoch: 50
+
+net:
+  _target_: src.models.components.sparc.SPARC
+  arch: ViT-L/14@336px
+  image_size: 336
+  temperature: 0.07                     # softmax
+  feature_map_idx: [5, 11, 17, 23]      # [0, 12, 23] [6, 12, 18] [5, 11, 17, 23] index of resnetblock in ViT
+  prompt_learner:
+    _target_: src.models.components.coop.AnomalyPromptLearner
+    _partial_: true
+    tokenizer:
+      _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+    prompt_length: 12                     # length of learnable prompts
+    context_length: 77                    # defaut 77 for openai clip
+    truncate: false
+    class_names: ["object"]
+    # class_names: ${prompt.class_names}
+    # state_template: ${prompt.state_template}
+    state_template: 
+      normal: ["{}"]
+      anomaly: ["damaged {}"]
+  text_encoder:
+    _target_: src.models.components.text_encoder.TextMapEncoder
+    _partial_: true
+  adapter:
+    # _target_: torch.nn.Linear
+    # in_features: 1024                   # clip vit feature dim, defaut 1024 for openai clip
+    # out_features: 1024
+    # bias: false
+    _target_: src.models.components.adapter.Linear
+    in_features: 1024                     # clip vit feature dim, defaut 1024 for openai clip
+    out_features: 1024
+    hidden_features: null                 # set null, same as nn.Linear
+    dropout_prob: 0.0
+    bias: false
+  fusion:
+    _target_: src.models.components.cross_modal.DotProductFusion
+  embedding_dim: 768                    # clip fusion featrue dim, default 768, only effective for non null
+
+loss:
+  cross_entropy:
+    _target_: torch.nn.CrossEntropyLoss
+  focal:
+    _target_: src.models.components.loss.FocalLoss
+  dice:
+    _target_: src.models.components.loss.BinaryDiceLoss
+
+k_shot: false
+
+filter: true
+
+enable_validation: false
+
+compile: false

SOWA/configs/model/sparc_prompt.yaml

@@ -0,0 +1,74 @@
+_target_: src.models.anomaly_clip_module.AnomalyCLIPModule
+
+optimizer:
+  _target_: torch.optim.AdamW
+  _partial_: true
+  lr: 0.001
+  weight_decay: 0.2
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  _partial_: true
+  mode: min
+  factor: 0.1
+  patience: 5
+
+# scheduler:
+#   _target_: src.models.components.scheduler.WarmupCosineAnnealingLR
+#   _partial_: true
+#   warmup_epochs: 10
+#   total_epoch: 50
+
+net:
+  _target_: src.models.components..sparc.SPARC
+  arch: ViT-L/14@336px
+  image_size: 336
+  temperature: 0.07                     # softmax
+  feature_map_idx: [5, 11, 17, 23]      # [0, 12, 23] [6, 12, 18] [5, 11, 17, 23] index of resnetblock in ViT
+  share_weight: true                    # whether the adapter shares weights for different feature maps
+  prompt_learner:
+    _target_: src.models.components.coop.PromptEncoder
+    _partial_: true
+    tokenizer:
+      _target_: src.models.components.clip.simple_tokenizer.SimpleTokenizer
+    context_length: 77                    # defaut 77 for openai clip
+    truncate: false
+    class_names: ${prompt.class_names}
+    prompt_normal: ${prompt.template.normal}
+    prompt_abnormal: ${prompt.template.abnormal}
+    prompt_templates: ${prompt.template.templates}
+  adapter:
+    _target_: torch.nn.Linear
+    in_features: 1024                   # clip vit feature dim, defaut 1024 for openai clip
+    out_features: 1024
+    bias: false
+    # _target_: src.models.components.adapter.BasicLayer
+    # _partial_: true
+    # input_resolution: [24, 24]          # (image_size - kerner_size) / stride + 1. eg. 24 = (224 - 14) / 14 + 1
+    # window_size: 12
+    # depth: 1                            # if depth < 2, thers is no window shift
+    # num_heads: 8
+    # hidden_features: null               # set null, same as nn.Linear
+    # cpb_dim: 64
+    # value_only: true
+    # drop: 0.0
+    # attn_drop: 0.1
+  fusion:
+    _target_: src.models.components.cross_modal.DotProductFusion
+  embedding_dim: 768                    # clip fusion featrue dim, default 768, only effective for non null
+
+loss:
+  cross_entropy:
+    _target_: torch.nn.CrossEntropyLoss
+  focal:
+    _target_: src.models.components.loss.FocalLoss
+  dice:
+    _target_: src.models.components.loss.BinaryDiceLoss
+
+k_shot: false
+
+filter: true
+
+enable_validation: false
+
+compile: false

SOWA/configs/paths/default.yaml

@@ -0,0 +1,18 @@
+# path to root directory
+# this requires PROJECT_ROOT environment variable to exist
+# you can replace it with "." if you want the root to be the current working directory
+root_dir: ${oc.env:PROJECT_ROOT}
+
+# path to data directory
+data_dir: ${paths.root_dir}/data/
+
+# path to logging directory
+log_dir: ${paths.root_dir}/logs/
+
+# path to output directory, created dynamically by hydra
+# path generation pattern is specified in `configs/hydra/default.yaml`
+# use it to store all files generated during the run, like ckpts and metrics
+output_dir: ${hydra:runtime.output_dir}
+
+# path to working directory
+work_dir: ${hydra:runtime.cwd}

SOWA/configs/prompt/default.yaml

@@ -0,0 +1,5 @@
+# configs/prompt/default.yaml
+defaults:
+  - object
+  - state_template
+  - template

SOWA/configs/prompt/object.yaml

@@ -0,0 +1,29 @@
+class_names:
+  - object
+  - pipe_fryum
+  - metal_nut
+  - pcb1
+  - tile
+  - screw
+  - pcb2
+  - wood
+  - zipper
+  - cable
+  - fryum
+  - pill
+  - capsule
+  - hazelnut
+  - pcb4
+  - leather
+  - bottle
+  - cashew
+  - macaroni2
+  - grid
+  - chewinggum
+  - transistor
+  - macaroni1
+  - candle
+  - capsules
+  - pcb3
+  - carpet
+  - toothbrush

SOWA/configs/prompt/state_template.yaml

@@ -0,0 +1,91 @@
+state_template:
+  anomaly:
+  - a photo of the damaged {}.
+  - a bright photo of the damaged {}.
+  - a dark photo of the damaged {}.
+  - a close-up photo of the damaged {}.
+  - a black and white photo of the damaged {}.
+  - a blurry photo of the damaged {}.
+  - a blurry photo of a damaged {}.
+  - a photo of the small damaged {}.
+  - a photo of the large damaged {}.
+  - there is a damaged {} in the scene.
+  - this is one damaged {} in the scene.
+  - a photo of the broken {}.
+  - a bright photo of the broken {}.
+  - a dark photo of the broken {}.
+  - a close-up photo of the broken {}.
+  - a black and white photo of the broken {}.
+  - a blurry photo of the broken {}.
+  - a blurry photo of a broken {}.
+  - a photo of the small broken {}.
+  - a photo of the large broken {}.
+  - there is a broken {} in the scene.
+  - this is one broken {} in the scene.
+  - a photo of the {} with flaw.
+  - a bright photo of the {} with flaw.
+  - a dark photo of the {} with flaw.
+  - a close-up photo of the {} with flaw.
+  - a black and white photo of the {} with flaw.
+  - a blurry photo of the {} with flaw.
+  - a blurry photo of a {} with flaw.
+  - a photo of the small {} with flaw.
+  - a photo of the large {} with flaw.
+  - there is a {} with flaw in the scene.
+  - this is one {} with flaw in the scene.
+  - a photo of the {} with defect.
+  - a bright photo of the {} with defect.
+  - a dark photo of the {} with defect.
+  - a close-up photo of the {} with defect.
+  - a black and white photo of the {} with defect.
+  - a blurry photo of the {} with defect.
+  - a blurry photo of a {} with defect.
+  - a photo of the small {} with defect.
+  - a photo of the large {} with defect.
+  - there is a {} with defect in the scene.
+  - this is one {} with defect in the scene.
+  normal:
+  - a photo of the {}.
+  - a bright photo of the {}.
+  - a dark photo of the {}.
+  - a close-up photo of the {}.
+  - a black and white photo of the {}.
+  - a blurry photo of the {}.
+  - a blurry photo of a {}.
+  - a photo of the small {}.
+  - a photo of the large {}.
+  - there is a {} in the scene.
+  - this is one {} in the scene.
+  - a photo of the flawless {}.
+  - a bright photo of the flawless {}.
+  - a dark photo of the flawless {}.
+  - a close-up photo of the flawless {}.
+  - a black and white photo of the flawless {}.
+  - a blurry photo of the flawless {}.
+  - a blurry photo of a flawless {}.
+  - a photo of the small flawless {}.
+  - a photo of the large flawless {}.
+  - there is a flawless {} in the scene.
+  - this is one flawless {} in the scene.
+  - a photo of the {} without flaw.
+  - a bright photo of the {} without flaw.
+  - a dark photo of the {} without flaw.
+  - a close-up photo of the {} without flaw.
+  - a black and white photo of the {} without flaw.
+  - a blurry photo of the {} without flaw.
+  - a blurry photo of a {} without flaw.
+  - a photo of the small {} without flaw.
+  - a photo of the large {} without flaw.
+  - there is a {} without flaw in the scene.
+  - this is one {} without flaw in the scene.
+  - a photo of the {} without defect.
+  - a bright photo of the {} without defect.
+  - a dark photo of the {} without defect.
+  - a close-up photo of the {} without defect.
+  - a black and white photo of the {} without defect.
+  - a blurry photo of the {} without defect.
+  - a blurry photo of a {} without defect.
+  - a photo of the small {} without defect.
+  - a photo of the large {} without defect.
+  - there is a {} without defect in the scene.
+  - this is one {} without defect in the scene.

SOWA/configs/prompt/template.yaml

@@ -0,0 +1,51 @@
+template:
+  normal:
+    - '{}'
+    - 'flawless {}'
+    - 'perfect {}'
+    - 'unblemished {}'
+    - '{} without flaw'
+    - '{} without defect'
+    - '{} without damage'
+  abnormal:
+    - 'damaged {}'
+    - 'broken {}'
+    - '{} with flaw'
+    - '{} with defect'
+    - '{} with damage'
+  templates:
+    - 'a bad photo of a {}.'
+    - 'a low resolution photo of the {}.'
+    - 'a bad photo of the {}.'
+    - 'a cropped photo of the {}.'
+    - 'a bright photo of a {}.'
+    - 'a dark photo of the {}.'
+    - 'a photo of my {}.'
+    - 'a photo of the cool {}.'
+    - 'a close-up photo of a {}.'
+    - 'a black and white photo of the {}.'
+    - 'a bright photo of the {}.'
+    - 'a cropped photo of a {}.'
+    - 'a jpeg corrupted photo of a {}.'
+    - 'a blurry photo of the {}.'
+    - 'a photo of the {}.'
+    - 'a good photo of the {}.'
+    - 'a photo of one {}.'
+    - 'a close-up photo of the {}.'
+    - 'a photo of a {}.'
+    - 'a low resolution photo of a {}.'
+    - 'a photo of a large {}.'
+    - 'a blurry photo of a {}.'
+    - 'a jpeg corrupted photo of the {}.'
+    - 'a good photo of a {}.'
+    - 'a photo of the small {}.'
+    - 'a photo of the large {}.'
+    - 'a black and white photo of a {}.'
+    - 'a dark photo of a {}.'
+    - 'a photo of a cool {}.'
+    - 'a photo of a small {}.'
+    - 'there is a {} in the scene.'
+    - 'there is the {} in the scene.'
+    - 'this is a {} in the scene.'
+    - 'this is the {} in the scene.'
+    - 'this is one {} in the scene.'

SOWA/configs/train.yaml

@@ -0,0 +1,52 @@
+# @package _global_
+
+# specify here default configuration
+# order of defaults determines the order in which configs override each other
+defaults:
+  - _self_
+  - data: anomaly_clip
+  - model: anomaly_clip
+  - callbacks: default
+  - logger: many_loggers # set logger here or use command line (e.g. `python train.py logger=tensorboard`)
+  - trainer: default
+  - paths: default
+  - extras: default
+  - hydra: default
+
+  # information of object and prompt template
+  - prompt: default
+
+  # experiment configs allow for version control of specific hyperparameters
+  # e.g. best hyperparameters for given model and datamodule
+  - experiment: null
+
+  # config for hyperparameter optimization
+  - hparams_search: null
+
+  # optional local config for machine/user specific settings
+  # it's optional since it doesn't need to exist and is excluded from version control
+  - optional local: default
+
+  # debugging config (enable through command line, e.g. `python train.py debug=default)
+  - debug: null
+
+# task name, determines output directory path
+task_name: "train"
+
+# tags to help you identify your experiments
+# you can overwrite this in experiment configs
+# overwrite from command line with `python train.py tags="[first_tag, second_tag]"`
+tags: ["dev"]
+
+# set False to skip model training
+train: True
+
+# evaluate on test set, using best model weights achieved during training
+# lightning chooses best weights based on the metric specified in checkpoint callback
+test: True
+
+# simply provide checkpoint path to resume training
+ckpt_path: null
+
+# seed for random number generators in pytorch, numpy and python.random
+seed: 2025