ahmdtaha / simsiam Goto Github PK

Hello, in your code, the dataset of cifar10 is used. But the file of cifar10 python is specialized. I want to use my own pictures to train, how do I prepare? Thank you very much.

I have a corpus that is not CIFAR. I am curious what other image sizes are supported?

Thanks for sharing the codes, Ahmed. Just to share what I got :)

2021-01-09 00:21:57,403 [INFO] train: Epoch: [798][96/97] Time 0.152 ( 0.166) Data 0.000 ( 0.006) Loss -0.960 (-0.945)
2021-01-09 00:22:04,898 [INFO] train: Finish the Trn Features -> Tst Features
2021-01-09 00:22:08,495 [INFO] train: Acc 88.69243421052632
2021-01-09 00:22:08,810 [INFO] train: ==> ETA: 0:00:54 GPU-M: 2712.0 GPU-U: 25
2021-01-09 00:22:08,810 [INFO] train: Overall Timing[798/1] epoch_time 0.4580 (0.4623) validation_time 0.1827 (0.1880) train_time 0.2701 (0.2606)
2021-01-09 00:22:09,491 [INFO] train: Epoch: [799][ 0/97] Time 0.679 ( 0.679) Data 0.587 ( 0.587) Loss -0.965 (-0.965)
2021-01-09 00:22:09,497 [INFO] train: Epoch: [799][ 0/97] Time 0.686 ( 0.686) Data 0.577 ( 0.577) Loss -0.954 (-0.954)
2021-01-09 00:22:09,508 [INFO] train: Epoch: [799][ 0/97] Time 0.697 ( 0.697) Data 0.573 ( 0.573) Loss -0.956 (-0.956)
2021-01-09 00:22:09,524 [INFO] train: Epoch: [799][ 0/97] Time 0.713 ( 0.713) Data 0.553 ( 0.553) Loss -0.959 (-0.959)
2021-01-09 00:22:24,815 [INFO] train: Epoch: [799][96/97] Time 0.152 ( 0.165) Data 0.000 ( 0.006) Loss -0.964 (-0.946)
2021-01-09 00:22:24,816 [INFO] train: Epoch: [799][96/97] Time 0.152 ( 0.165) Data 0.000 ( 0.006) Loss -0.940 (-0.946)
2021-01-09 00:22:24,833 [INFO] train: Epoch: [799][96/97] Time 0.155 ( 0.165) Data 0.000 ( 0.006) Loss -0.953 (-0.946)
2021-01-09 00:22:24,837 [INFO] train: Epoch: [799][96/97] Time 0.161 ( 0.165) Data 0.000 ( 0.007) Loss -0.949 (-0.945)
2021-01-09 00:22:32,206 [INFO] train: Finish the Trn Features -> Tst Features
2021-01-09 00:22:35,884 [INFO] train: Acc 88.7952302631579
2021-01-09 00:22:36,190 [INFO] train: ==> ETA: 0:00:27 GPU-M: 2712.0 GPU-U: 31
2021-01-09 00:22:36,190 [INFO] train: Overall Timing[799/1] epoch_time 0.4563 (0.4623) validation_time 0.1830 (0.1880) train_time 0.2682 (0.2606)

I noticed this repo did not have a requirements.txt file, so I've created a pull request to add one: #5

Thanks for sharing the code！
I try to train on cifar10 and find some problem: gpu_id and process rank are not match

        self.sampler = torch.utils.data.distributed.DistributedSampler(
            trn_dataset, rank=cfg.gpu, num_replicas=cfg.world_size, shuffle=True
        )

Thanks for your implementation, could you provide the accuracy you got with cifar100?

After running classifer_main, I obtained the following folders.

Among them, there is a file named model_best.pth. How can I use this file? Or what is the purpose of the classifier_main file? I haven't quite understood.