Skip to content

vghost2008/wml1

Repository files navigation

#WML

Machine learning tools library.

##Dependencies

  • Linux or Mac OS
  • Python ≥ 3.6
  • scipy
  • yacs
  • OpenCV
  • pycocotools
  • gcc & g++ ≥ 4.9
  • tensorflow ≥ 1.10

##Installation

pip install -r requirements.txt
cd tfop
make clean
make -j18

##Prepare datasets

###COCO

  1. Extract COCO datasets into the following stracture:
coco
├── annotations
├── val2017
└── train2017
  1. Create tf record for object detection, instance segmentation and semantic segmentation
python datasets_tools/create_coco_tf_record.py --data_dir ~/ai/mldata/coco/
  1. Create tf record for 2D pose estimation
python datasets_tools/create_coco_tf_kp_record.py --data_dir ~/ai/mldata/coco

###Generate other datasets

The usage is similar to COCO datasets toolkit.

  • Pascal VOC: datasets_tools/create_pascal_voc_tf_record.py
  • MOT: datasets_tools/create_mot_tf_record.py
  • LabelMe: datasets_tools/create_labelme_tf_record.py

##Change the default buildin datasets location

You can set location for buildin datasets by modify the value of object_detection2/data/datasets/buildin.py:dataset_root_path

##Train

###Train RetinaNet

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/coco/RetinaNetN101.yaml --gpus 0 1 2 3 4 5 6 7

###Train Mask-RCNN

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/coco/Mask-RCNN-FPN-seph.yaml --gpus 0 1 2 3 4 5 6 7

###Train Cascade Mask-RCNN

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/coco/Cascade-Mask-RCNN-FPN-N.yaml --gpus 0 1 2 3 4 5 6 7

###Train OpenPose

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/keypoints/OpenPose-coco.yaml --gpus 0 1 2 3 4 5 6 7

###Train HRNet

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/keypoints/OpenPose-coco.yaml --gpus 0 1 2 3 4 5 6 7

###Train FairMOT

python object_detection_tools/train_net_on_multi_gpus.py --config-file object_detection2/default_configs/MOT/FairMOT.yaml --gpus 0 1 2 3 4 5 6 7

###Eval

example

python object_detection_tools/eval_net.py --config-file object_detection2/default_configs/coco/RetinaNet.yaml --gpus 0

##Predict on images

example

python object_detection_tools/predict_on_images.py --test_data_dir ../test_imgs --config-file object_detection2/default_configs/coco/RetinaNet.yaml

##MOT Track

python object_detection_tools/mot_track.py --config-file object_detection2/default_configs/MOT/FairMOT.yaml --gpus 0

License

WML itself is released under the MIT License (refer to the LICENSE file for details).

##Authors

    Wang Jie  bluetornado@zju.edu.cn

    Copyright 2017 The WML Authors.  All rights reserved.