EasyClassification (easycls) is a classification framework based on PyTorch.
A deep-learning classification framework with clean code and fine extensibility.
Main features:
- CPU and GPU(CUDA) support
- YAML configuration file for all functions and parameters
- auto log system
-
python >= 3.6
-
pytorch
-
pyyaml
Configurate sections in config.yml:
- basic: basic train/eval settings
- data: dataset and path
- model: model architecture and parameters
- loss: loss function and parameters
- optimizer: optimizer on model parameters
- lr_scheduler: learning rate scheduler (usually learning rate decay)
- speed: computing speed related settings
Train a new model:
python tools\train.py -c config.yml Resume training model with a previous checkpoint:
python tools\train.py -c config.yml -r <name>.ptEval a model with a checkpoint:
python tools\eval.py -c config.yml -r <name>.pt