Skip to content

Regarding the issue of achieving only 0.64 accuracy on the HMDB51 dataset for 1-shot 5-way evaluation? #8

Open
@ssp789

Description

@ssp789

First of all, thank you very much for your dedication to this project. I am very interested in your project.

I am running it on a machine with two 1080ti GPUs. Regarding the settings, I have changed QUERY_PER_CLASS from the default 5 to 1, BATCH_SIZE to 2, BATCH_SIZE_PER_TASK to 1, and NUM_GPUS to 2.

The ending epoch for training is
[06/18 12:42:55][INFO] utils.logging: 88: {"_type": "train_iter", "epoch": "8/300", "eta": "2442 days, 6:42:01", "iter": "5000/500000", "loss": 1.968674, "lr": 0.000000, "time_diff": 1.440410, "top1_err": 0.000000, "top5_err": 0.000000}
[06/18 12:42:55][INFO] utils.logging: 88: {"RAM": "13.10/125.50 GB", "_type": "train_epoch", "epoch": "8/300", "eta": "833 days, 19:57:43", "gpu_mem": "8.39 GB", "loss": 2.024753, "lr": 0.000000, "time_diff": 0.493488, "top1_err": 4.358000, "top5_err": 0.000000}

The ending epoch for validation is
[06/18 12:37:29][INFO] utils.logging: 88: {"_type": "val_iter", "epoch": "16/300", "eta": "0:00:00", "gpu_mem": "8.39 GB", "iter": "5000/5000", "time_diff": 0.418341, "top1_err": 39.999996, "top5_err": 0.000000}
[06/18 12:37:30][INFO] utils.logging: 88: {"RAM": "17.17/125.50 GB", "_type": "val_epoch", "epoch": "16/300", "gpu_mem": "8.39 GB", "min_top1_err": 31.265998, "min_top5_err": 0.000000, "time_diff": 0.427307, "top1_err": 35.943999, "top5_err": 0.000000}

The final result during the testing phase is
[06/18 15:22:46][INFO] utils.logging: 88: {"_type": "val_iter", "epoch": "1/300", "eta": "0:00:14", "gpu_mem": "0.82 GB", "iter": "4950/5000", "time_diff": 0.294500, "top1_err": 39.999996, "top5_err": 0.000000}
[06/18 15:23:13][INFO] utils.logging: 88: {"_type": "val_iter", "epoch": "1/300", "eta": "0:00:00", "gpu_mem": "0.82 GB", "iter": "5000/5000", "time_diff": 1.051394, "top1_err": 39.999996, "top5_err": 0.000000}
[06/18 15:23:14][INFO] utils.logging: 88: {"RAM": "13.12/125.50 GB", "_type": "val_epoch", "epoch": "1/300", "gpu_mem": "0.82 GB", "min_top1_err": 35.511999, "min_top5_err": 0.000000, "time_diff": 0.433443, "top1_err": 35.511999, "top5_err": 0.000000}
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 6.0, acc: 0.7971303308090872
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 1.0, acc: 0.36624848239579116
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 9.0, acc: 0.7836443032949583
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 2.0, acc: 0.6622222222222223
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 7.0, acc: 0.3078462770216173
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 5.0, acc: 0.9492868462757528
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 0.0, acc: 0.8161822466614297
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 8.0, acc: 0.8520900321543409
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 3.0, acc: 0.223468161794153
[06/18 15:23:14][INFO] test_net_few_shot: 207: class: 4.0, acc: 0.680145572179539

Is this process normal? I'm looking forward to your reply. Thank you very much.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions