I refer matterport/Mask_RCNN, and modify code to TensorFlow 2.1 version.
Also I implementation of model on MS COCO dataset.
I use Anaconda to create environment and code edit on Spyder
- tensorflow 2.1.0 (2.3.0 also can run, but will very many warning)
- Keras 2.3.1
- others in requirements.txt
- OS: Windows 10 64bit
- GPU: NVIDIA GTX 1060
- CUDA: 10.1
- CuDNN: 7.6.5 for CUDA 10.1
Update: 2020 Dec. 12
Only input and evaluate COCO data by pre-trained model.
For COCO training work not yet.
- run 'mrcnn' on TF2 (use Shapes data)
- input COCO dataset for training
- run inspect code from matterport/Mask_RCNN/samples/coco
- check training happen on COCO dataset
- remove mask part for implementation of Faster RCNN model (Here)