AndreaPi
commented
5 years ago @VellalaVineethKumar I'm not one of the repository developers, but I'd say that depends on what's your goal. I guess you want to train on COCO...otherwise you don't need COCO at all, you can just download the pretrained COCO weights from here:
https://github.com/matterport/Mask_RCNN/releases
If you want to train on COCO, then you need (at least)
http://images.cocodataset.org/zips/train2014.zip http://images.cocodataset.org/zips/val2014.zip http://images.cocodataset.org/annotations/annotations_trainval2014.zip
To run some of the notebooks, you'll also need the 5K minival and the 35K validation-minus-minival subsets, as described here:
https://github.com/matterport/Mask_RCNN#ms-coco-requirements
My suggestion: once you downloaded everything and decompressed in the right folders, start by running
https://github.com/matterport/Mask_RCNN/blob/master/samples/coco/inspect_data.ipynb
https://github.com/matterport/Mask_RCNN/blob/master/samples/coco/inspect_model.ipynb
before going full monty and trying to train Mask R-CNN on COCO. Also, make sure you've got a beefy GPU: I don't think that training this baby on COCO is something you do overnight. It takes about an hour to train it on the much simpler balloon dataset, with these instances
from the official coco website
Images 2014 Train images [83K/13GB] 2014 Val images [41K/6GB] 2014 Test images [41K/6GB] 2015 Test images [81K/12GB] 2017 Train images [118K/18GB] 2017 Val images [5K/1GB] 2017 Test images [41K/6GB] 2017 Unlabeled images [123K/19GB] or 2014 Train/ValDetection 2015, Captioning 2015, Detection 2016, Keypoints 2016 2014 Testing Captioning 2015 2015 Testing Detection 2015, Detection 2016, Keypoints 2016 2017 Train/Val/Test Detection 2017, Keypoints 2017, Stuff 2017, Detection 2018, Keypoints 2018, Stuff 2018, Panoptic 2018 2017 Unlabeled[optional data for any competition] which dataset should i download?