accv_tardis-pose
Science Score: 44.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
✓CITATION.cff file
Found CITATION.cff file -
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (6.5%) to scientific vocabulary
Repository
Basic Info
- Host: GitHub
- Owner: ACCV2024-Paper356
- License: apache-2.0
- Language: Python
- Default Branch: main
- Size: 14.2 MB
Statistics
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
TARDIS-Pose: Targeted Distillation of Self-Supervised ViT features for Animal Pose Estimation
Requirements
- Linux (not tested on other platforms)
- Python 3.6 or newer
- PyTorch >= 0.4.0
- CUDA >= 9.0
- cuDNN >= 7.1
- mmcv >= 2.0.1
- mmengine >= 0.8.4
Getting started
Install dependencies
pip install -r requirements.txt
Prepare datasets
Download AP-10K from: https://github.com/AlexTheBad/AP-10K
Extract and place data in ./datasets/animal_data/AP-10K
Alternatively, create a symbolic link pointing to your dataset location.
Training
Extract ViT features
To extract features using DINOv2-Large for AP-10K training set:
python scripts/extract_dino.py
Extracted features will be saved to ./data/dino
Run distillation
Run distillation of HRNet-w32 on AP-10K:
python scripts/train.py configs/animal_2d_keypoint/dinopose/ap10k/distill_hrnet_ap10k-256x256.py
Run distillation of ResNet-50 on AP-10K:
python scripts/train.py configs/animal_2d_keypoint/dinopose/ap10k/distill_res50_ap10k-256x256.py
Models and logs will be saved to ./work_dirs.
Train keypoint detection
Update path to distilled model in config file if necessary.
Train fully supervised:
python scripts/train.py configs/animal_2d_keypoint/dinopose/ap10k/supervised_distill_hrnet_ap10k-256x256.py
Train few-shot
python scripts/train.py configs/animal_2d_keypoint/dinopose/ap10k/fewshot<n_imgs>_distill_hrnet_ap10k-256x256.py
For example, to train an HRNet on AP-10K with 5 labeled images per animal:
python scripts/train.py configs/animal_2d_keypoint/dinopose/ap10k/fewshot05_distill_hrnet_ap10k-256x256.py
Owner
- Login: ACCV2024-Paper356
- Kind: user
- Repositories: 1
- Profile: https://github.com/ACCV2024-Paper356
Citation (CITATION.cff)
cff-version: 1.2.0 message: "If you use this software, please cite it as below." authors: - name: "MMPose Contributors" title: "OpenMMLab Pose Estimation Toolbox and Benchmark" date-released: 2020-08-31 url: "https://github.com/open-mmlab/mmpose" license: Apache-2.0
GitHub Events
Total
Last Year
Dependencies
- pytorch/pytorch ${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel build
- pytorch/pytorch ${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel build
- loguru ==0.6.0
- numpy ==1.21.6
- onnxruntime ==1.14.1
- onnxruntime-gpu ==1.8.1
- albumentations >=0.3.2
- numpy *
- torch >=1.8
- docutils ==0.16.0
- markdown *
- myst-parser *
- sphinx ==4.5.0
- sphinx_copybutton *
- sphinx_markdown_tables *
- urllib3 <2.0.0
- mmcv >=2.0.0,<2.2.0
- mmdet >=3.0.0,<3.3.0
- mmengine >=0.4.0,<1.0.0
- requests *
- shapely ==1.8.4
- mmcv >=2.0.0rc4
- mmengine >=0.6.0,<1.0.0
- munkres *
- regex *
- scipy *
- titlecase *
- torch >1.6
- torchvision *
- xtcocotools >=1.13
- chumpy *
- json_tricks *
- matplotlib *
- munkres *
- numpy *
- opencv-python *
- pillow *
- scipy *
- torchvision *
- xtcocotools >=1.12
- coverage * test
- flake8 * test
- interrogate * test
- isort ==4.3.21 test
- parameterized * test
- pytest * test
- pytest-runner * test
- xdoctest >=0.10.0 test
- yapf * test
- addict *
- albumentations *
- h5py *
- joblib *
- mmcv *
- mmdet *
- mmengine *
- mmpose *
- pycocotools *
- seaborn *
- setuptools *
- six *