Daniel Bolya

`S` is not score but an extra semantic segmentation loss. That going down is perfectly normal (and is what should be happening). As for lr, I think what you have...

Not sure if this is the issue, but if you're using multiple GPUs, you should use a batch size that's evenly divisible by the number of GPUs you're using. The...

1. Your new backbone doesn't need DCN, just remember to change the backbone in the yolact++ config. 2. `init_backbone` takes in the image-net pretrained weights file and loads it into...

@Auth0rM0rgan The most important parameter would be `selected_layers`. The rest, you should probably keep the same (unless you want to tune these yourself). `selected_layers` determines which backbone layers we should...

@ahkarami It's a good thing to have, so I'll add it to my TODO list (along with efficientnet).

@Auth0rM0rgan I'll add it to the list but this is starting to be too much now. @elfpattern ¯\\\_(ツ)\_/¯

You'll have to implement that yourself, but masks are created here: https://github.com/dbolya/yolact/blob/02bde378e98cd7e2831639ff782c1f5a5046771a/eval.py#L155 All eval options use the same code to generate and display masks, so implement your logging code here...

1. You might not be giving your GPU enough work. Try increasing the batch size and image size and see what happens. 2. The results in the paper were done...

Hi, thanks for the suggestion! I actually tried something similar in the [paper](https://arxiv.org/pdf/2303.17604.pdf):  It did improve the quantitative numbers somewhat, but I found the small gain in FID was...

Thanks for writing this! Do you know what the speed difference might be between this and the native pytorch solution (w/o onnx)? Essentially, this is a native pytorch implementation of...