Ting Chen
Ting Chen
> How to convert coco annotations into target sequences? (especially those with more than one polygon) we use a separator to indicate different polygon.s > Do the starting point and...
you can offset logits before sampling if you want to disable certain predictions (e.g. set -1e9 for logits that correspond to class/coordinates). but we find that just free sampling is...
this should be possible, though i haven't tried anything like that. see https://arxiv.org/pdf/2111.10050.pdf
feel free to share your rep here if you made it work eventually!
One workaround is to reshape the tensor from (bsz, instances, seqlen) into (bsz * instances, seqlen) for the model and then reshape back after done. Hope this helps. On Tue,...
This looks like some data issue as the complaint was about a keyerror probably related to image id. On Wed, Oct 12, 2022 at 1:03 AM ross-Hr ***@***.***> wrote: >...
this looks like the checkpoint specified (either pretrained checkpoint, or checkpoint restored from last training in the same model directory) is different from the configured architecture/encoder, please check if the...
you should be able to do pdb in the code when running in eager mode
this is mainly to avoid shortcuts in learning. For example, once the model confirms a fake/noise object appears, all the rest objects can be easily recognized as fake/noise. On Sat,...
Thanks for pointing this out. Both training and test image size are 224x224 in simclr repo (and the paper). It is possible using 256x256 could improve (slightly).