search

LiheYoung / Depth-Anything

[CVPR 2024] Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data. Foundation Model for Monocular Depth Estimation
https://depth-anything.github.io
Apache License 2.0
7.01k stars 539 forks source link

BDD100K: 8M Images? #194

Closed ShenZheng2000 closed 5 months ago

ShenZheng2000 commented 5 months ago

According to Table 1 (screenshot below), the model was trained on 8 million unlabeled images from BDD100K. However, the downloadable BDD100K dataset contains only 100,000 images.

Therefore, I wonder if the authors sample these 8 million images from video clips. If so, please provide the list of videos and the sampling rate. Thanks!

image

LiheYoung commented 5 months ago

Yes, we evenly sample 8M images from the raw videos.

ShenZheng2000 commented 5 months ago

Could you please specify the sampling rate (extract 1 frame from how many frames)?

LiheYoung commented 5 months ago

Our sampling rate is one image / 10 frames. We sample them from 70K training videos of BDD100K.

ShenZheng2000 commented 5 months ago

Thanks.