README.md

Accelerated DL R&D

PyTorch framework for Deep Learning research and development. It was developed with a focus on reproducibility, fast experimentation and code/ideas reusing. Being able to research/develop something new, rather than write another regular train loop.
Break the cycle - use the Catalyst!

Project manifest. Part of PyTorch Ecosystem. Part of Catalyst Ecosystem:

• Alchemy - Experiments logging & visualization
• Catalyst - Accelerated Deep Learning Research and Development
• Reaction - Convenient Deep Learning models serving

Catalyst at AI Landscape.

---

Catalyst.Video

You will learn how to build video classification pipeline with transfer learning using the Catalyst framework to get reproducible results.

Preparation

You have to split your video files into classes in the following format:

video_dataset/
    action_0/
        video_abc.avi
        ...
    action_1/
        video_abcd.avi
        ...

Data preprocessing

catalyst-data tag2label \
    --in-dir=./data/video_dataset \
    --out-dataset=./data/dataset.csv \
    --out-labeling=./data/labeling.json \
    --tag-column=video

python tsn/process_data.py \
    --in-csv=./data/dataset.csv \
    --datapath=./data/video_dataset \
    --out-csv=./data/dataset_processed.csv \
    --out-dir=./data/video_dataset_processed \
    --n-cpu=4 \
    --verbose

Data splitting

PYTHONPATH=tsn python tsn/prepare_splits.py \
    --base_path=./data/video_dataset \
    --dataset_path=./data/dataset.csv \
    --out_folder=./data \
    --k_fold=5 \
    --use_folds=4

Model training

CUDA_VISIBLE_DEVICES=0 catalyst-dl run --config configs/train.yml

Tensorboard

CUDA_VISIBLE_DEVICES="" tensorboard --logdir /mnt/ssd1/logs/tsn --host=0.0.0.0 --port=6006

WIP

Model inference

Don't forget to fix data_params at infer.yaml

export LOGDIR=/mnt/ssd1/logs/tsn
CUDA_VISIBLE_DEVICES=1 PYTHONPATH=tsn catalyst-dl infer \
   --expdir=tsn \
   --resume=${LOGDIR}/checkpoint.best.pth.tar \
   --out-prefix=${LOGDIR}/dataset.predictions.npy \
   --config=${LOGDIR}/config.json,./configs/infer.yml \
   --verbose

Docker (TODO)

Build

docker build -t tsn .

Run

docker run --rm -it \
    -p 8893:8893 \
    --ipc=host \
    --runtime=nvidia \
    tsn \
    flask run -p 8893 --host=0.0.0.0