search

serizba / SfM-TTR

SfM-TTR: Using Structure from Motion for Test-Time Refinement of Single-View Depth Networks
GNU General Public License v3.0
23 stars 1 forks source link

how to process my own images #1

Open whulc opened 1 year ago

whulc commented 1 year ago

I have taken some photos and use colmap to reconstruct them. How can I use the code to generate the depthmap？

serizba commented 1 year ago

Hi @whulc ,

You should create a dataset class for your images, like kitti.py. Then you can create a TunerDataset using your dataset and the path to the reconstructed COLMAP models.

From there you can finetune an SfMTuner.

whulc commented 1 year ago

Is ground truth depth map necessary for the finetune and prediction? It seems that in kitti.py, the gt depthmap is provided or calculated from lidar pointcloud.

serizba commented 1 year ago

Hi! No, ground truth depth map is only used for evaluation and is not used at all during the finetuning. You can just return a dummy placeholder if you don't have gt available.

I will soon update the repo with a version that does not require to return the gt.

Thanks!