deep_sort_pytorch
deep_sort_pytorch copied to clipboard
ZQPei
high loss in validation when I train
I download the market1501 dataset and reorganize the train and test dataset, but when I train the model, the train accuracy increase but the test accuracy is alwanys very low, Epoch : 40 [progress:9.9%]time:5.85s Loss:0.19229 Correct:1227/1280 Acc:95.859% [progress:19.7%]time:12.49s Loss:0.19317 Correct:2455/2560 Acc:95.898% [progress:29.6%]time:6.21s Loss:0.16215 Correct:3698/3840 Acc:96.302% [progress:39.4%]time:6.23s Loss:0.18000 Correct:4937/5120 Acc:96.426% [progress:49.3%]time:6.22s Loss:0.18456 Correct:6169/6400 Acc:96.391% [progress:59.1%]time:6.27s Loss:0.18482 Correct:7395/7680 Acc:96.289% [progress:69.0%]time:6.28s Loss:0.19110 Correct:8616/8960 Acc:96.161% [progress:78.8%]time:6.00s Loss:0.17664 Correct:9854/10240 Acc:96.230% [progress:88.7%]time:6.27s Loss:0.16078 Correct:11090/11520 Acc:96.267% [progress:98.5%]time:6.28s Loss:0.17974 Correct:12330/12800 Acc:96.328% Testing ... [progress:100.0%]time:41.70s Loss:13.60742 Correct:24/15913 Acc:0.151% Learning rate adjusted to 0.0010000000000000002 anyone can sovle this?
im having the same problem im trying to train it on my own dataset of cars .. any solu!
你好，请问训练时除了改输入数据集路径还要改啥吗？我也下了1501的数据集，改了输入路径，但是程序报错：RuntimeError: Found 0 files in subfolders of: /Users/sunyuhua/Desktop/Market1501/bounding_box_train
Supported extensions are: .jpg,.jpeg,.png,.ppm,.bmp,.pgm,.tif,.tiff,.webp
请大神赐教
@DerekSunYH you should divide the Market1501 to different folders such as 0000,0001,....
@yustaub is there any other repo that you know where i can train the tracker to track cars ..?
When I download the MARS dataset, I saw that train and test dataset have different classes. Solution--> The train and test dataset must have the same classes.
@joaquinCaceres yes you are right for market 1501 dataset we need to split train dataset into train and validation. not to use test set as validation.
