search

AI-Chen / HRNet-V2

HRNet 的Tensorflow实现
MIT License
34 stars 11 forks source link

It's a little difficult to understand the code（line 95-97）in HighResolutionModule.py #2

Open kubl88 opened 5 years ago

kubl88 commented 5 years ago 
        for k in range(len(temp_layer)):
            fuse = 0
            fuse = fuse + temp_layer[k]

Please have a look. Thanks

minogame commented 4 years ago

The implementation of the _make_fuse_layers is totally incorrect, please check the official repo.

Linfeng-Tang commented 4 years ago

I don't think it handles downsampling well, and according to the official REPORT method, it should downsample i-j times when j<i (i.e. perform i-j conv(3×3, stride=2)), but in its implementation it only performs once