sir,I find it(fcn8-vgg16) accuracy only 0.2657, how the result so terrible ?

IanTaehoonYoo / semantic-segmentation-pytorch

Pytorch implementation of FCN, UNet, PSPNet, and various encoder models.

MIT License

86 stars 19 forks source link

sir,I find it(fcn8-vgg16) accuracy only 0.2657, how the result so terrible ? #12

Closed LeeNannaner closed 3 years ago

LeeNannaner commented 3 years ago

And the test result is lower greatly than the train.

LeeNannaner commented 3 years ago

Dear sir, could you give me some suggestions?

IanTaehoonYoo commented 3 years ago

Hi LeeNananer,

I guess you trained the sample dataset in this project, and it is too little data to train. It is called 'overfitting', so you should download the Cityspace dataset. Also, I suggest the image size should be greater than 500x500 if you train PSPnet.

LeeNannaner commented 3 years ago

Oh，oh，thanks sir. I will have a try.

zhou-rui1 commented 3 years ago

Hi, how can it train with the image size greater than 500x500 ?

once I try to train with greater size (batchsize=2), it give this, : CUDA out of memory. Tried to allocate 18.00 MiB (GPU 0; 15.90 GiB total capacity; 15.02 GiB already allocated; 17.75 MiB free; 15.08 GiB reserved in total by PyTorch)

Appreciate for any guidance!