Pano

Guidance to run the code

1. follow the instruction on how to download coco panoptic dataset and store in the root folder as coco/ following this tree structure

coco
├── annotations
├── converted_data
├── __MACOSX
├── panoptic_annotations_trainval2017.zip
├── test2017
├── train2017
└── val2017

2. Follow the instruction and install CocoApi

3. in the root folder type the following command to generate change panoptic format to detection for trainval/val2017

python2 scripts/panoptic2detection_coco_format.py   --input_json_file coco/annotations/panoptic_train2017.json \
--output_json_file coco/converted_data/panoptic_coco_detection_train2017.json \ 
 --segmentations_folder coco/annotations/panoptic_train2017/ --categories_json_file panopticapi/panoptic_coco_categories.json

Repeat the same for segmentation format change

python2 scripts/panoptic2segmentation_coco_format.py   --input_json_file coco/annotations/panoptic_train2017.json \
--output_json_file coco/converted_data/panoptic_segmentation_train2017.json \  
 --segmentations_folder coco/annotations/panoptic_train2017/ --categories_json_file panopticapi/panoptic_coco_categories.json

4. Install requirements for Mask-RCNN

pip install -r requirement.txt

5. Run the following command to start training

PYTHONPATH=./ python samples/coco/coco2017.py train --dataset=coco/ --model=imagenet # Python3