search

HRNet / HRNet-Object-Detection

Object detection with multi-level representations generated from deep high-resolution representation learning (HRNetV2h). This is an official implementation for our TPAMI paper "Deep High-Resolution Representation Learning for Visual Recognition". https://arxiv.org/abs/1908.07919
Apache License 2.0
644 stars 97 forks source link

Question on Upsample #44

Closed liuhengyue closed 4 years ago

liuhengyue commented 4 years ago

Hi, thanks for the great work. I do have a question on the upsample method. In the paper on mult-scale feature fusion, the upsample is "bilinear" followed by the 1x1 conv, but in the implementation I saw the Upsample kwarg is "nearest". Do I miss or miss-understand anything here? Btw, will it influence the feature representation much (based on your intuition or experiments).

wondervictor commented 4 years ago

@liuhengyue Hi, the upsampling in the backbone network (HRNet) is nearest. The influence here is neglectable.