Skip to content

Good iou in evaluation but poor performance in reality #79

@Echo990

Description

@Echo990

Hi,

I have been trying to train a grasp model based on your work and would appreciate your help.

I experimented with two different datasets. The first dataset consisted of around 300 images of plastic vegetable toys against a black background, and I obtained pretty good results. This model also performs well in real-world scenarios.

However, when I moved on to the second dataset, which primarily includes building blocks and dolls (the dolls are almost spherical), the model achieves a high Intersection over Union (IoU) during evaluation but does not perform well in real-life situations.

I also attempted to incorporate additional augmentation methods, such as adding Gaussian noise and applying blur to the RGB images. While these techniques helped stabilize the training process and maintained a good IoU in evaluation, the real-world performance remains subpar.

Could you please provide some suggestions on how to improve the model's performance in real scenarios?

Thank you!

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