This is an variant version of MS-G3D
The difference is
- Distributed Training
- SE module
- Complement Graph and Full Graph
- Dense Padding
- Python >= 3.7
- PyTorch >= 1.5.1
- NVIDIA Apex (auto mixed precision training)
- PyYAML, tqdm, tensorboardX
- The general training template command:
python3 -u -m torch.distributed.launch --nproc_per_node=8 main_dist \
--config <config file> \
--work-dir <place to keep things (weights, checkpoints, logs)> \
[--base-lr <base learning rate>] \
[--batch-size <batch size>] \
[--weight-decay <weight decay>] \
[--forward-batch-size <batch size during forward pass, useful if using only 1 GPU>]
- The general testing template command:
python3 -u main.py
--config <config file>
--work-dir <place to keep things>
--device <GPU IDs to use>
--weights <path to model weights>
[--test-batch-size <...>]
- Template for joint-bone two-stream fusion:
You can build a list of scores generated at test stage, and create a file list such as:
ensemble.txt
path/to/first/model/test/score.pkl 1
path/to/second/model/test/score.pkl 1
...
path/to/last/model/test/score.pkl 1
cd ensemble
python3 ensemble.py \
--ground-truth <label_file> \
--method <softmax / sum> \
--output <output fiel to save merged score> \
--result-list <models list to be merged>
- Resume training from checkpoint
python3 -u -m torch.distributed.launch --nproc_per_node=8 main_dist \
--config <config file> \
--work-dir <place to keep things (weights, checkpoints, logs)> \
--checkpoint <checkpoint to be resumed> \
[--base-lr <base learning rate>] \
[--batch-size <batch size>] \
[--weight-decay <weight decay>] \
[--forward-batch-size <batch size during forward pass, useful if using only 1 GPU>]
This repo is based on
Thanks to the original authors for their work!
## Contact
Please email `[email protected]` for further questions
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent