daitranskku/mmdetection-distiller

This is a knowledge distillation toolbox based on mmdetection.

If you want to distill model in OpenMMLab related repos, please use MMRazor!!

If you are intrested in KD，you also could contact me by Wechat, and I will invite you to the KD group.

This project is based on mmdetection(v-2.9.0), all the usage is the same as mmdetection including training , test and so on.

Distiller Zoo

• Channel-wise Distillation for Semantic Segmentation
• Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors

Installation

• Set up a new conda environment: conda create -n distiller python=3.7

• Install pytorch

• Install mmcv ( 1.2.4 <= mmcv-full < 1.3 )

• Install mmdetection-distiller

  git clone https://github.com/pppppM/mmdetection-distiller.git
  cd mmdetection-distiller
  pip install -r requirements/build.txt
  pip install -v -e .

Train

#single GPU
python tools/train.py configs/distillers/cwd/cwd_retina_rx101_64x4d_distill_retina_r50_fpn_2x_coco.py

#multi GPU
bash tools/dist_train.sh configs/distillers/cwd/cwd_retina_rx101_64x4d_distill_retina_r50_fpn_2x_coco.py 8

Test

#single GPU
python tools/test.py configs/distillers/cwd/cwd_retina_rx101_64x4d_distill_retina_r50_fpn_2x_coco.py $CHECKPOINT --eval bbox

#multi GPU
bash tools/dist_train.sh configs/distillers/cwd/cwd_retina_rx101_64x4d_distill_retina_r50_fpn_2x_coco.py $CHECKPOINT 8 --eval bbox

Lisence

This project is released under the Apache 2.0 license.