C-3-Framework
C-3-Framework copied to clipboard
dramatic result in CSRNet_UCF50 model
I conducted the experiment on CSRNet_UCF50 model, the res is : mae:196.5/mse:235.3, which is not match with mae:266.1/mse:397.5 in https://github.com/gjy3035/Awesome-Crowd-Counting#SPN.
C3F's implementation is not official, which adopts different training parameters and scheme. The latter (266.1/397.5) is the original result in CSR paper.
Thanks!And i found the vgg-decoder network train on shanghaiB dataset, the pred is always 0 during the train and the test.
You may need to use our provided parameters.
@yuzehui1996 Hi, did you solve the pred problem for vgg-decoder in SHHB?