Could you share your instances_trainval35k.json.zip file ?

[FostorHUNT]

No description provided.

[ellisbrown]

here they are on S3: https://s3.amazonaws.com/amdegroot-datasets/instances_trainval35k.json.zip

We also have created a script to form the trainval35k dataset for you on the COCO branch. We hope to merge these into master soon
https://github.com/amdegroot/ssd.pytorch/blob/coco/data/scripts/COCO2014.sh

[FostorHUNT]

Sorry @ellisbrown , I knew this S3 link before adding the issue. Unfortunately, I got 'access denied' response when using the command line
curl -LO https://s3.amazonaws.com/amdegroot-datasets/instances_trainval35k.json.zip
The file I got is:
<?xml version="1.0" encoding="UTF-8"?> <Error><Code>AccessDenied</Code><Message>Access Denied</Message><RequestId>A4DF808B97BA75E5</RequestId><HostId>L88xkRSVyyagBYgVHKb860AkBPFdH1RcYZK1kfOaJWILmVe3YLeFEUyU9v6xPvbiOuL05yJUUUo=</HostId></Error>

[amdegroot]

Sorry about that, I still need to make the bucket public.

This link should work. It's the same annotations, just taken from Ross Girshick's py-faster-rcnn repo, which is where I pulled ours from.

[ujsyehao]

@amdegroot Hi, this is valminusminival2014.json, it just contains val35k annotations, and it doesn't contain train80k annotations.
I am confused about it. Do you use tain2014.json first then use valminusminival2014.json later? I don't find the corresponding code in the project.