Faster-rcnn.pytorch: Does it really work on coco? can anyone write a step to step actions to make it work please????

[isalirezag]

I am having a hard time to make it work for coco dataset...

I tried both coco 2014 and coco 2017. BUT it sounds like thre exist the same problem...

Can anyone explain how to do it?

I constantly receive error regarding not having instances_valminusminival2017.json'

in trainval.py we have:
elif args.dataset == "coco": args.imdb_name = "coco_2017_train+coco_2017_valminusminival" args.imdbval_name = "coco_2017_minival"
but when you download and make the coco dataset there is no coco_2017_valminusminival or coco_2017_train...

can anyone PLEASE explain how to make or build or download:
coco_2017_valminusminival and coco_2017_minival
or
coco_2014_valminusminival and coco_2014_minival

It would be a hugee help :

I have seen more that 50 issues regarding this problem for coco

and please dont say it is explain in this link. because if it was clear we werent asking again :)

any help...

[isalirezag]

I still dont know how to create those files and if someone know i will appreciate it if they can provide a detailed explanation for it....

Anyways, i found that for coco2014 we can download them from here:
https://drive.google.com/drive/folders/0B1_fAEgxdnvJSmF3YUlZcHFqWTQ

[isalirezag]

also i found this useful:
https://github.com/facebookresearch/Detectron/tree/master/detectron/datasets/data