search

RaymondWang987 / VDW_Dataset_Toolkits

The official generation code and toolkits of VDW dataset (ICCV 2023)
Other
35 stars 11 forks source link

A question about the Invalid Sample Filtering #3

Open Jiang01151144 opened 1 month ago

Jiang01151144 commented 1 month ago

Thank you very much for sharing the code to generate stereo disparity data. By running your scripts, we successfully obtained a small dataset for our project.

Actually, I have a question regarding the so-called “Invalid Sample Filtering” part. What’s the purpose for removing samples with “average range of horizontal disparity is less than 15 pixels” as in your instructions？I find that it will remove lots of videos in our scenarios.

Thanks again and looking forward to your reply.

RaymondWang987 commented 1 month ago

Thank you very much for sharing the code to generate stereo disparity data. By running your scripts, we successfully obtained a small dataset for our project.

Actually, I have a question regarding the so-called “Invalid Sample Filtering” part. What’s the purpose for removing samples with “average range of horizontal disparity is less than 15 pixels” as in your instructions？I find that it will remove lots of videos in our scenarios.

Thanks again and looking forward to your reply.

Glad to hear that!

The restriction of horizontal disparity is conducted for the quality of ground truth. For instance, if the range of horizontal disparity is too small (<15), the disparity map will be noisy after saving with uint-16 (converting values<15 to 65535). This operation is also described in the MiDaS paper. The threshold 15 can be adjusted based on your data and scenes, e.g., MiDaS adopts 10 pixels. You can visualize your disparity maps and adjust the threshold accordingly. Thanks!