A problem of incomplete network structure of backbone

[Sean-Wick]

In model.py，for the depth-dominant branch of backbone, is it missing sparsed_feature0_plus = torch.cat([rgb_feature0_plus, sparsed_feature], 1) ？ Between 183 and 184.

[JUGGHM]

Thanks for your interest. We previously did not concat them because (1) It was a mistake which should have been corrected. (2) The concatenation led to higher computational cost which we could not afford at that time. However, in recent experiments we found that such a concatenation is harmful, and we roughly assumed one possible reason: it could impair the bottleneck effect of the connection from CD branch to DD branch via CD depth (?).

[Sean-Wick]

Ok, thanks for your answer !