I wanna reproduce ARTrack-L-384

[XJLeoYu]

but seems that the files you provided didn't match,it report bug: dim 1024 should be divided by num_heads 12

[AlexDotHam]

Sorry, the num_heads should be 16, i will change that soon.

[XJLeoYu]

When I changed the num_heads back to 16 in config.py,it's still bugy

[AlexDotHam]

I have updated the YAML, on my server, and it is executing normally (it should be noticed that I only have 3090 for training, so I reset the batch_size to 2). Maybe there is something wrong with your operation. If you have any other questions, you can mail me and add my WeChat.

[XJLeoYu]

I was just planing to inference not traning,not pretty sure how to find you mail or wechat account on github

[AlexDotHam]

You can mail me yfbai@stu.xjtu.edu.cn

[Jasper0420]

您好，我在使用384给出的权重推理时出现以下shape不匹配的问题？

size mismatch for pix_head.decoder.layers.3.multihead_attn.in_proj_bias: copying a param with shape torch.Size([3072]) from checkpoint, the shape in current model is torch.Size([2304]).
        size mismatch for pix_head.decoder.layers.3.multihead_attn.out_proj.weight: copying a param with shape torch.Size([1024, 1024]) from checkpoint, the shape in current model is torch.Size([768, 768]).
        size mismatch for pix_head.decoder.layers.3.multihead_attn.out_proj.bias: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([768]).
        size mismatch for pix_head.decoder.layers.3.norm_3.weight: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([768]).
        size mismatch for pix_head.decoder.layers.3.norm_3.bias: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([768]).
        size mismatch for pix_head.decoder.layers.3.mlpz.fc1.weight: copying a param with shape torch.Size([4096, 1024]) from checkpoint, the shape in current model is torch.Size([3072, 768]).
        size mismatch for pix_head.decoder.layers.3.mlpz.fc1.bias: copying a param with shape torch.Size([4096]) from checkpoint, the shape in current model is torch.Size([3072]).
        size mismatch for pix_head.decoder.layers.3.mlpz.fc2.weight: copying a param with shape torch.Size([1024, 4096]) from checkpoint, the shape in current model is torch.Size([768, 3072]).

请问应该如何修改yaml文件或者其他配置吗？

[AlexDotHam]

It seems like you use the ARTrack-L-384 model checkpoints with a ARTrack-B-384 yaml, you can check the yaml you use to match the checkpoints.

[Jasper0420]

Thanks! Good luck!