search

Pongpisit-Thanasutives / Variations-of-SFANet-for-Crowd-Counting

The official implementation of "Encoder-Decoder Based Convolutional Neural Networks with Multi-Scale-Aware Modules for Crowd Counting"
https://ieeexplore.ieee.org/document/9413286
GNU General Public License v3.0
110 stars 32 forks source link

train detail for shanghaiA and B #18

Open 1286710929 opened 3 years ago

1286710929 commented 3 years ago

I train the datasets of shanghai A and B. But I find that the result （MAE） of SFANet is 68.02 and 10.32 for A and B. The process of the experiment is completely in accordance with your article. And I use the model from UCF data set for pre-training

knightyxp commented 3 years ago

the published code do not have CANet branch, not coordinated with the paper report, the baseline of SFANet is 59 according to my exp, however, when I add ASPP(means using the M-SFANet model according to the author's code) the SHA MAE is only 61, ridiculous, the res in the paper can not be reproduced (i do not know whether the reviewer of ICPR know this thing)