---
task_categories:
- image-segmentation
- object-detection
- robotics
- zero-shot-object-detection
size_categories:
- n>1T
configs:
- config_name: MegaPose-ShapeNetCore
data_files: MegaPose-ShapeNetCore/*.tar
- config_name: MegaPose-GSO
data_files: MegaPose-GSO/*.tar
---
# BOP: Benchmark for 6D Object Pose Estimation
The goal of BOP is to capture the state of the art in estimating the 6D pose, i.e. 3D translation and 3D rotation, of rigid objects from RGB/RGB-D images. An accurate, fast, robust, scalable and easy-to-train method that solves this task will have a big impact in application fields such as robotics or augmented reality.
Homepage: https://bop.felk.cvut.cz/home/
Toolkit: https://github.com/thodan/bop_toolkit
## Downloading datasets
#### Option 1: Using `huggingface_hub`:
Click to expand
a. Install the library:
```
pip install --upgrade huggingface_hub
```
b. Download the dataset:
```
from huggingface_hub import snapshot_download
dataset_name = "hope"
local_dir = "./datasets"
snapshot_download(repo_id="bop-benchmark/datasets",
allow_patterns=f"{dataset_name}/*zip",
repo_type="dataset",
local_dir=local_dir)
```
If you want to download the entire BOP datasets (~3TB), please remove the `allow_patterns` argument. More options are available in the [official documentation](https://huggingface.co/docs/huggingface_hub/main/en/guides/download).
Click to expand
a. Install the library:
```
pip install -U "huggingface_hub[cli]"
```
b. Download the dataset:
```
export LOCAL_DIR=./datasets
export DATASET_NAME=hope
huggingface-cli download bop-benchmark/datasets --include "$DATASET_NAME/*.zip" --local-dir $LOCAL_DIR --repo-type=dataset
```
Please remove this argument `--include "$DATASET_NAME/*.zip"` to download entire BOP datasets (~3TB). More options are available in the [official documentation](https://huggingface.co/docs/huggingface_hub/main/en/guides/download).
Click to expand
Similar `wget` command as in [BOP website](https://bop.felk.cvut.cz/datasets/) can be used to download the dataset from huggingface hub:
```
export SRC=https://huggingface.co/datasets/bop-benchmark/datasets/resolve/main
wget $SRC/lm/lm_base.zip # Base archive
wget $SRC/lm/lm_models.zip # 3D object models
wget $SRC/lm/lm_test_all.zip # All test images ("_bop19" for a subset)
wget $SRC/lm/lm_train_pbr.zip # PBR training images
```
Click to expand
a. Install the library:
```
pip install -U "huggingface_hub[cli]"
```
b. Log-in and create a token
```
huggingface-cli login
```
Then go to [this link](https://huggingface.co/settings/tokens) and generate a token. IMPORTANT: the token should have write access as shown below:
Make sure you are in the bop-benchmark group by running:
```
huggingface-cli whoami
```
c. Upload dataset:
The command is applied for both folders and specific files:
```
# Usage: huggingface-cli upload bop-benchmark/datasets [local_path] [path_in_repo] --repo-type=dataset --create-pr
```
For example, to upload hope dataset:
```
export LOCAL_FOLDER=./datasets/hope
export HF_FOLDER=/hope
huggingface-cli upload bop-benchmark/datasets $LOCAL_FOLDER $HF_FOLDER --repo-type=dataset --create-pr
```
Click to expand
a. Install the library:
```
pip install --upgrade huggingface_hub
```
b. Creating a pull-request:
We recommend organizing the dataset in a folder and then uploading it to the huggingface hub. For example, to upload `lmo`:
```
from huggingface_hub import HfApi, CommitOperationAdd
dataset_name = "lmo"
local_dir = "./datasets/lmo"
operations = []
for file in local_dir.glob("*"):
add_commit = CommitOperationAdd(
path_in_repo=f"/{dataset_name}",
path_or_fileobj=local_dir,
)
operations.append(add_commit)
api = HfApi()
MY_TOKEN = # get from https://huggingface.co/settings/tokens
api.create_commit(repo_id="bop-benchmark/datasets",
repo_type="dataset",
commit_message=f"adding {dataset_name} dataset",
token=MY_TOKEN,
operations=operations,
create_pr=True)
```
If your dataset is large (> 500 GB), you can upload it in chunks by adding the `multi_commits=True, multi_commits_verbose=True,` argument. More options are available in the [official documentation](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/package_reference/hf_api#huggingface_hub.HfApi.create_pull_request).