train loss

[SophieChang66]

hello, have you trained the model directly on the shanghaitech dataset, or have you trained it on imagenet then fineturn it ??? I directly train it , but the loss is about is 60, I doubt it has something wrong

[miao0913]

Hi, the result presented in the paper was trained from scratch, however we also tried with finetuning from Imagenet classification task, which gives almost the same result. You should be careful with the data format in caffe NCHW, so better visual either your input or output. Thanks 2018-06-20 6:46 GMT+02:00 chengtingting <notifications@github.com>:

…

hello, have you trained the model directly on the shanghaitech dataset, or have you trained it on imagenet then fineturn it ??? I directly train it , but the loss is about is 60, I doubt it has something wrong — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9>, or mute the thread <https://github.com/notifications/unsubscribe-auth/APmD0Zq74o6Y8q-tNKNWV7qK0EvUqhaVks5t-dOegaJpZM4UumjV> .

-- Dr. Miaojing Shi Research Scientist LINKMEDIA - Inria Rennes Email: miaojing.shi@inria.fr CC: miao0913@gmail.com Web: https://sites.google.com/site/miaojingshi/