JiawangBian / sc_depth_pl

SC-Depth (V1, V2, and V3) for Unsupervised Monocular Depth Estimation Webpage:https://jiawangbian.github.io/sc_depth_pl/
GNU General Public License v3.0
445 stars 71 forks source link

About Subsample video frames #38

Open wjlszhang opened 1 year ago

wjlszhang commented 1 year ago

Hi, @JiawangBian

I get my own data by a blender script, the basic idea is to take the pictures while the camera moves around the object, thus I get ~100 pics in each scense (take one pic when the camer moves center degrees). Before the training, I use below to generate valid frames

python generate_valid_frame_index.py --dataset_dir $DATASET

However, I only get 1 pic in each scense, something wrong when I take the pic?

The pic looks like below:

无标题

JiawangBian commented 1 year ago

You may not need to generate a valid frame index. You can just set a frame sequence, like 5 or 3 FPS

wjlszhang commented 1 year ago

Thanks @JiawangBian, I'll do like as you suggested. Another question do you think the abs_rel could be down to 0.01 if all my scense (both train and val) are generated with the same tooth with similar camera poses? Thanks much!