Reproducing lamp scene paper results
Hi, I tried to reproduce the lamp scene results from the paper using the provided “example_scenes” dataset. The lamp dataset contains images in the folders:
- hdr
- ldr
- ldr_true
However, the ldr images appear to contain only one exposure variant. Are the other images with multiple exposure variants missing?
Also, I managed to create a mesh with sdfstudio using the provided “lamp_sdfstudio” dataset, but the result is different from the 3D mesh provided under “extra_data/3d_assets/lamp/mesh_simplified”.
My result:
Only the bottom part actually contains the lamp scene when I zoom in. Did you get similar results? I would be grateful if you could let me know what post-processing steps you did in Blender, e.g. cropped the mesh to the bottom part, decimated the mesh or edited the normals?
Then I tried the script for emission optimization. However, the script expects a file “transformed_cameras.json” which is not supplied, only “transforms.json” is supplied. Based on the code in “optimize.py”, these files do not have the same format. Is there a script to create the file “transformed_cameras.json”? It would also already help if you could provide the structure of the file.
Thank you very much for your efforts and Keep up the good work! :)