mandelakori commited on
Commit
e547693
1 Parent(s): d5850ce

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +32 -91
README.md CHANGED
@@ -1,118 +1,59 @@
1
  ---
2
- license: apache-2.0
3
  tags:
4
  - object-detection
5
- - vision
6
- datasets:
7
- - coco
8
- widget:
9
- - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/savanna.jpg
10
- example_title: Savanna
11
- - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/football-match.jpg
12
- example_title: Football Match
13
- - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/airport.jpg
14
- example_title: Airport
15
  ---
16
 
17
- # DETR (End-to-End Object Detection) model with ResNet-50 backbone
18
 
19
- DEtection TRansformer (DETR) model trained end-to-end on COCO 2017 object detection (118k annotated images). It was introduced in the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Carion et al. and first released in [this repository](https://github.com/facebookresearch/detr).
20
 
21
- Disclaimer: The team releasing DETR did not write a model card for this model so this model card has been written by the Hugging Face team.
22
 
23
- ## Model description
24
 
25
- The DETR model is an encoder-decoder transformer with a convolutional backbone. Two heads are added on top of the decoder outputs in order to perform object detection: a linear layer for the class labels and a MLP (multi-layer perceptron) for the bounding boxes. The model uses so-called object queries to detect objects in an image. Each object query looks for a particular object in the image. For COCO, the number of object queries is set to 100.
 
 
 
26
 
27
- The model is trained using a "bipartite matching loss": one compares the predicted classes + bounding boxes of each of the N = 100 object queries to the ground truth annotations, padded up to the same length N (so if an image only contains 4 objects, 96 annotations will just have a "no object" as class and "no bounding box" as bounding box). The Hungarian matching algorithm is used to create an optimal one-to-one mapping between each of the N queries and each of the N annotations. Next, standard cross-entropy (for the classes) and a linear combination of the L1 and generalized IoU loss (for the bounding boxes) are used to optimize the parameters of the model.
28
 
29
- ![model image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/detr_architecture.png)
30
 
31
- ## Intended uses & limitations
32
 
33
- You can use the raw model for object detection. See the [model hub](https://huggingface.co/models?search=facebook/detr) to look for all available DETR models.
34
 
35
- ### How to use
36
 
37
- Here is how to use this model:
 
38
 
39
- ```python
40
- from transformers import DetrImageProcessor, DetrForObjectDetection
41
- import torch
42
- from PIL import Image
43
- import requests
44
 
45
- url = "http://images.cocodataset.org/val2017/000000039769.jpg"
46
- image = Image.open(requests.get(url, stream=True).raw)
47
 
48
- # you can specify the revision tag if you don't want the timm dependency
49
- processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
50
- model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
51
 
52
- inputs = processor(images=image, return_tensors="pt")
53
- outputs = model(**inputs)
54
 
55
- # convert outputs (bounding boxes and class logits) to COCO API
56
- # let's only keep detections with score > 0.9
57
- target_sizes = torch.tensor([image.size[::-1]])
58
- results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]
59
 
60
- for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
61
- box = [round(i, 2) for i in box.tolist()]
62
- print(
63
- f"Detected {model.config.id2label[label.item()]} with confidence "
64
- f"{round(score.item(), 3)} at location {box}"
65
- )
66
- ```
67
- This should output:
68
- ```
69
- Detected remote with confidence 0.998 at location [40.16, 70.81, 175.55, 117.98]
70
- Detected remote with confidence 0.996 at location [333.24, 72.55, 368.33, 187.66]
71
- Detected couch with confidence 0.995 at location [-0.02, 1.15, 639.73, 473.76]
72
- Detected cat with confidence 0.999 at location [13.24, 52.05, 314.02, 470.93]
73
- Detected cat with confidence 0.999 at location [345.4, 23.85, 640.37, 368.72]
74
- ```
75
 
76
- Currently, both the feature extractor and model support PyTorch.
77
 
78
- ## Training data
79
 
80
- The DETR model was trained on [COCO 2017 object detection](https://cocodataset.org/#download), a dataset consisting of 118k/5k annotated images for training/validation respectively.
 
 
81
 
82
- ## Training procedure
83
 
84
- ### Preprocessing
85
 
86
- The exact details of preprocessing of images during training/validation can be found [here](https://github.com/google-research/vision_transformer/blob/master/vit_jax/input_pipeline.py).
87
-
88
- Images are resized/rescaled such that the shortest side is at least 800 pixels and the largest side at most 1333 pixels, and normalized across the RGB channels with the ImageNet mean (0.485, 0.456, 0.406) and standard deviation (0.229, 0.224, 0.225).
89
-
90
- ### Training
91
-
92
- The model was trained for 300 epochs on 16 V100 GPUs. This takes 3 days, with 4 images per GPU (hence a total batch size of 64).
93
-
94
- ## Evaluation results
95
-
96
- This model achieves an AP (average precision) of **42.0** on COCO 2017 validation. For more details regarding evaluation results, we refer to table 1 of the original paper.
97
- ### BibTeX entry and citation info
98
-
99
- ```bibtex
100
- @article{DBLP:journals/corr/abs-2005-12872,
101
- author = {Nicolas Carion and
102
- Francisco Massa and
103
- Gabriel Synnaeve and
104
- Nicolas Usunier and
105
- Alexander Kirillov and
106
- Sergey Zagoruyko},
107
- title = {End-to-End Object Detection with Transformers},
108
- journal = {CoRR},
109
- volume = {abs/2005.12872},
110
- year = {2020},
111
- url = {https://arxiv.org/abs/2005.12872},
112
- archivePrefix = {arXiv},
113
- eprint = {2005.12872},
114
- timestamp = {Thu, 28 May 2020 17:38:09 +0200},
115
- biburl = {https://dblp.org/rec/journals/corr/abs-2005-12872.bib},
116
- bibsource = {dblp computer science bibliography, https://dblp.org}
117
- }
118
- ```
 
1
  ---
 
2
  tags:
3
  - object-detection
4
+ - computer-vision
5
+ language:
6
+ - en
7
+ pipeline_tag: object-detection
 
 
 
 
 
 
8
  ---
9
 
10
+ # AISAK-Visual
11
 
12
+ ## Overview:
13
 
14
+ AISAK-Visual, part of the AISAK system, is a pretrained model for image captioning based on the BLIP framework. Altered by the AISAK team from the https://huggingface.co/Salesforce/blip-image-captioning-large model, this model utilizes a ViT base backbone for unified vision-language understanding and generation.
15
 
16
+ ## Model Information:
17
 
18
+ - **Model Name**: AISAK-Visual
19
+ - **Version**: 2.0
20
+ - **Model Architecture**: Transformer with ViT base backbone
21
+ - **Specialization**: AISAK-Visual is part of the broader AISAK system and is specialized in image captioning tasks.
22
 
23
+ ## Intended Use:
24
 
25
+ AISAK-Visual, as part of AISAK, is designed to provide accurate and contextually relevant captions for images. Whether used for conditional or unconditional image captioning tasks, AISAK-Visual offers strong performance across various vision-language understanding and generation tasks.
26
 
27
+ ## Performance:
28
 
29
+ AISAK-Visual, based on the BLIP framework, achieves state-of-the-art results on image captioning tasks, including image-text retrieval, image captioning, and VQA. Its generalization ability is demonstrated by its strong performance on video-language tasks in a zero-shot manner.
30
 
31
+ ## Ethical Considerations:
32
 
33
+ - **Bias Mitigation**: Efforts have been made to mitigate bias during training; however, users are encouraged to remain vigilant about potential biases in the model's output.
34
+ - **Fair Use**: Users should exercise caution when using AISAK-Visual in sensitive contexts and ensure fair and ethical use of the generated image captions.
35
 
36
+ ## Limitations:
 
 
 
 
37
 
38
+ - While AISAK-Visual demonstrates proficiency in image captioning tasks, it may not be suitable for tasks requiring domain-specific knowledge.
39
+ - Performance may vary when presented with highly specialized or out-of-domain images.
40
 
41
+ ## Deployment:
 
 
42
 
43
+ Inferencing for AISAK-Visual will be handled as part of the full deployment of the AISAK system in the future. The process is lengthy and intensive in many areas, emphasizing the goal of achieving the optimal system rather than the quickest. However, work is being done as fast as humanly possible. Updates will be provided as frequently as possible.
 
44
 
 
 
 
 
45
 
46
+ ## Caveats:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
47
 
48
+ - Users should verify important decisions based on AISAK-Visual's image captions, particularly in critical or high-stakes scenarios.
49
 
50
+ ## Model Card Information:
51
 
52
+ - **Model Card Created**: February 1, 2024
53
+ - **Last Updated**: February 19, 2024
54
+ - **Contact Information**: For any inquiries or communication regarding AISAK, please contact me at mandelakorilogan@gmail.com.
55
 
 
56
 
57
+ **© 2024 Mandela Logan. All rights reserved.**
58
 
59
+ No part of this model may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the copyright holder. Users are expressly prohibited from creating replications or spaces derived from this model, whether in whole or in part, without the explicit authorization of the copyright holder. Unauthorized use or reproduction of this model is strictly prohibited by copyright law.