search

hisfog / SfMNeXt-Impl

[AAAI 2024] Official implementation of "SQLdepth: Generalizable Self-Supervised Fine-Structured Monocular Depth Estimation", and more.
MIT License
91 stars 12 forks source link

Pretrained_pose weights #14

Open zsz-pro opened 1 year ago

zsz-pro commented 1 year ago

Could you please provide the pretrained_pose weights? I used my own trained weights to visualize the warped image, and it should be similar to the source frame, but it is actually more similar to the reference frame. I would like to check if this might be due to a poorly trained PoseNet

hisfog commented 1 year ago

https://drive.google.com/file/d/1gZNixqmUKIRIfbWYt3LPiGAg0CufpgBr/view?usp=drive_link

zsz-pro commented 1 year ago

Thanks, I have submitted the request！

hisfog commented 1 year ago

Thanks, I have submitted the request！

It should be a public link now

zsz-pro commented 1 year ago

Using the pre-trained PoseNet weights you provided, the image alignment has improved significantly, as seen in areas highlighted by the green box. However, there are still some alignment issues, as indicated by the red box. I'm curious about the significance of pose prediction in influencing the results because it seems that there is relatively limited work focused on improving PoseNet. Could you provide some insights? image