Cascade RCNN

This work builds on tf-eager-fasterrcnn

Faster R-CNN R-101-FPN model was implemented with TensorFlow2.0 Eager Execution.

Cascade RCNN model was implemented with TensorFlow2.0 Eager Execution.

Requirements

• Cuda 10.0
• Python 3.5
• TensorFlow 2.0.0
• cv2

Usage

see train_cascade_rcnn.ipynb, train_faster_rcnn.ipynb, inspect_model.ipynb and eval_model.ipynb

Download trained Faster R-CNN

• 百度网盘
• Google Drive

Make COCO2017 directory

Make your directory as follow.

COCO2017

--tmp_xml

--labelimg2coco.py

And put your images and xml files in tmp_xml.

Then just run labelimg2coco.py.

ToDO

• Muti-Scaling Training
• Pseudo Labeling
• GHM-C loss
• GHM-R loss
• Statistic Analysis of Dataset
• WBF
• TTA
• CutOut
• MixUp
• Dilated Conv
• SAG

Acknowledgement

This work builds on many excellent works, which include:

• matterport/Mask_RCNN
• open-mmlab/mmdetection