Skip to content

train result #4

Description

@ycyg8
RDVS (E↑) RDVS (S↑) RDVS (F↑) RDVS (M↓) ViDSOD-100 (E↑) ViDSOD-100 (S↑) ViDSOD-100 (F↑) ViDSOD-100 (M↓) DViSal (E↑) DViSal (S↑) DViSal (F↑) DViSal (M↓)
论文结果 0.913 0.879 0.827 0.026 0.918 0.894 0.868 0.020 0.914 0.840 0.818 0.046
权重评估 0.913 0.880 0.827 0.026 0.924 0.894 0.868 0.020 0.911 0.843 0.832 0.042
自己训练 0.883 0.853 0.786 0.026 0.829 0.769 0.663 0.050 0.908 0.841 0.815 0.045

The evaluation results in the second row were obtained using the weights file provided by the author, while the third row shows the results I got after training and evaluating myself based on the parameters in the paper and code. Should the parameter settings for these three datasets be the same? I also trained on a 3090. By checking the folder with the output saliency files provided by the author, I can see that the test sets are completely identical, so we can infer that the training sets are also the same, which confuses me. If you need any files from me, feel free to contact me. I hope to get an answer, thanks a lot.

These are the parameter settings
iters=2000,
sam2_config='configs/sam2.1/sam2.1_hiera_l.yaml',
sam2_ckpt='checkpoints/sam2.1_hiera_large.pt',
batch_size=1,
n_gpu=1,
local_rank=0,
base_lr=0.0001,
warmup=False,
warmup_period=10,
logpath=None,
frame_length=10,
num_frame_queries=30,
num_video_queries=8,
enable_memory=True

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions