Accelerated DL & RL

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.Segmentation Segmentation_Tests/statusIcon.svg" alt="Build Status">

Note: this repo uses advanced Catalyst Config API and could be a bit out-of-day right now.
Use Catalyst’s minimal examples section for a starting point and up-to-day use cases, please.

You will learn how to build image segmentation pipeline with transfer learning using the Catalyst framework.

Goals

Install requirements
Prepare data
Run: raw data → production-ready model
Get results
Customize own pipeline

1. Install requirements

Using local environment:

pip install -r requirements/requirements.txt

Using docker:

This creates a build catalyst-segmentation with the necessary libraries:

make docker-build

2. Get Dataset

Try on open datasets

You can use one of the open datasets

bash export DATASET="isbi" rm -rf data/ mkdir -p data if [[ "$DATASET" == "isbi" ]]; then # binary segmentation # http://brainiac2.mit.edu/isbi_challenge/ download-gdrive 1uyPb9WI0t2qMKIqOjFKMv1EtfQ5FAVEI isbi_cleared_191107.tar.gz tar -xf isbi_cleared_191107.tar.gz &>/dev/null mv isbi_cleared_191107 ./data/origin elif [[ "$DATASET" == "voc2012" ]]; then # semantic segmentation # http://host.robots.ox.ac.uk/pascal/VOC/voc2012/ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar tar -xf VOCtrainval_11-May-2012.tar &>/dev/null mkdir -p ./data/origin/images/; mv VOCdevkit/VOC2012/JPEGImages/* $_ mkdir -p ./data/origin/raw_masks; mv VOCdevkit/VOC2012/SegmentationClass/* $_ fi

Use your own dataset

Prepare your dataset

#### Data structure

Make sure, that final folder with data has the required structure:
bash /path/to/your_dataset/ images/ image_1 image_2 ... image_N raw_masks/ mask_1 mask_2 ... mask_N

#### Data location

The easiest way is to move your data:
```bash
mv /path/to/your_dataset/ /catalyst.segmentation/data/origin
In that way you can run pipeline with default settings. * If you prefer leave data in `/path/to/your_dataset/` * In local environment: * Link directorybash
ln -s /path/to/your_dataset $(pwd)/data/origin
* Or just set path to your dataset `DATADIR=/path/to/your_dataset` when you start the pipeline. * Using docker You need to set:bash
-v /path/to/your_dataset:/data \ #instead default $(pwd)/data/origin:/data
```
in the script below to start the pipeline.

3. Segmentation pipeline

Fast&Furious: raw data → production-ready model

The pipeline will automatically guide you from raw data to the production-ready model.

We will initialize Unet model with a pre-trained ResNet-18 encoder. During current pipeline model will be trained sequentially in two stages.

Binary segmentation pipeline

#### Run in local environment:

bash CUDA_VISIBLE_DEVICES=0 \ CUDNN_BENCHMARK="True" \ CUDNN_DETERMINISTIC="True" \ WORKDIR=./logs \ DATADIR=./data/origin \ IMAGE_SIZE=256 \ CONFIG_TEMPLATE=./configs/templates/binary.yml \ NUM_WORKERS=4 \ BATCH_SIZE=256 \ bash ./bin/catalyst-binary-segmentation-pipeline.sh

#### Run in docker:

bash export LOGDIR=$(pwd)/logs docker run -it --rm --shm-size 8G --runtime=nvidia \ -v $(pwd):/workspace/ \ -v $LOGDIR:/logdir/ \ -v $(pwd)/data/origin:/data \ -e "CUDA_VISIBLE_DEVICES=0" \ -e "USE_WANDB=1" \ -e "LOGDIR=/logdir" \ -e "CUDNN_BENCHMARK='True'" \ -e "CUDNN_DETERMINISTIC='True'" \ -e "WORKDIR=/logdir" \ -e "DATADIR=/data" \ -e "IMAGE_SIZE=256" \ -e "CONFIG_TEMPLATE=./configs/templates/binary.yml" \ -e "NUM_WORKERS=4" \ -e "BATCH_SIZE=256" \ catalyst-segmentation ./bin/catalyst-binary-segmentation-pipeline.sh

Semantic segmentation pipeline

#### Run in local environment:

bash CUDA_VISIBLE_DEVICES=0 \ CUDNN_BENCHMARK="True" \ CUDNN_DETERMINISTIC="True" \ WORKDIR=./logs \ DATADIR=./data/origin \ IMAGE_SIZE=256 \ CONFIG_TEMPLATE=./configs/templates/semantic.yml \ NUM_WORKERS=4 \ BATCH_SIZE=256 \ bash ./bin/catalyst-semantic-segmentation-pipeline.sh

#### Run in docker:

bash export LOGDIR=$(pwd)/logs docker run -it --rm --shm-size 8G --runtime=nvidia \ -v $(pwd):/workspace/ \ -v $LOGDIR:/logdir/ \ -v $(pwd)/data/origin:/data \ -e "CUDA_VISIBLE_DEVICES=0" \ -e "USE_WANDB=1" \ -e "LOGDIR=/logdir" \ -e "CUDNN_BENCHMARK='True'" \ -e "CUDNN_DETERMINISTIC='True'" \ -e "WORKDIR=/logdir" \ -e "DATADIR=/data" \ -e "IMAGE_SIZE=256" \ -e "CONFIG_TEMPLATE=./configs/templates/semantic.yml" \ -e "NUM_WORKERS=4" \ -e "BATCH_SIZE=256" \ catalyst-segmentation ./bin/catalyst-semantic-segmentation-pipeline.sh

The pipeline is running and you don’t have to do anything else, it remains to wait for the best model!

Visualizations

You can use W&B account for visualisation right after pip install wandb:

wandb: (1) Create a W&B account
wandb: (2) Use an existing W&B account
wandb: (3) Don't visualize my results

w&b binary segmentation metrics

Tensorboard also can be used for visualisation:

tensorboard --logdir=/catalyst.segmentation/logs

tf binary segmentation metrics

4. Results

All results of all experiments can be found locally in WORKDIR, by default catalyst.segmentation/logs. Results of experiment, for instance catalyst.segmentation/logs/logdir-191107-094627-2f31d790, contain:

checkpoints

The directory contains all checkpoints: best, last, also of all stages.
best.pth and last.pht can be also found in the corresponding experiment in your W&B account.

configs

The directory contains experiment`s configs for reproducibility.

logs

The directory contains all logs of experiment.
Metrics also logs can be displayed in the corresponding experiment in your W&B account.

code

The directory contains code on which calculations were performed. This is necessary for complete reproducibility.

5. Customize own pipeline

For your future experiments framework provides powerful configs allow to optimize configuration of the whole pipeline of segmentation in a controlled and reproducible way.

Configure your experiments

Common settings of stages of training and model parameters can be found in catalyst.segmentation/configs/_common.yml. model_params: detailed configuration of models, including:
model, for instance ResnetUnet detailed architecture description
using pretrained model stages: you can configure training or inference in several stages with different hyperparameters. In our example:
optimizer params first learn the head(s), then train the whole network

The CONFIG_TEMPLATE with other experiment`s hyperparameters, such as data_params and is here: catalyst.segmentation/configs/templates/binary.yml. The config allows you to define: data_params: path, batch size, num of workers and so on
* callbacks_params: Callbacks are used to execute code during training, for example, to get metrics or save checkpoints. Catalyst provide wide variety of helpful callbacks also you can use custom.

You can find much more options for configuring experiments in catalyst documentation.