Regarding the setting of c2wh = dict([(64,56), (128,28), (256,14) ,(512,7)]) in the fcanet.py file.

[Maxwell Wang]

In my model, the output feature map shape is (512, 16, 16), but I am worried that the adaptive_avg_pool2d() operation in the layer.py file will cause information loss. So I want to ask if the parameter: c2wh = dict([(64,56), ( 128,28), (256,14) ,(512,7)]) need to be changed?

[Maxwell Wang]

Change according to the corresponding size of the channel.

[cfzd]

@Max-Well-Wang
First, the adaptive_avg_pool2d operation should not harm the performance since our model works well in all COCO experiments. The input size on COCO is much bigger than ImageNet.

Second, if you want to change the size, you can simply modify the c2wh dict. Make the item (512, 7) to (512, 16) should work.

[Maxwell Wang]

@Max-Well-Wang
First, the adaptive_avg_pool2d operation should not harm the performance since our model works well in all COCO experiments. The input size on COCO is much bigger than ImageNet.

Second, if you want to change the size, you can simply modify the c2wh dict. Make the item (512, 7) to (512, 16) should work.

OK, thank you very much for your reply!

[Lewis0427]

请问top里面的最优频率分量 是不是只适用于77，一旦改变尺寸为1616 最优频率分量是否需要更换，还是仍起作用？