search

ashawkey / stable-dreamfusion

Text-to-3D & Image-to-3D & Mesh Exportation with NeRF + Diffusion.
Apache License 2.0
7.99k stars 710 forks source link

How can I did better in car model. #316

Open 523997931 opened 1 year ago

523997931 commented 1 year ago

Hello, this days I try to use multiview pictures to build the car model, I use the example data, and I also search some picture from websites, same as the example, I use the front, left, right, back, to reconstruct the 3d model ,but I don't get good results, either the example and my own data. image here is the example result without refine, image and this is my own data result, I try to adjust whte zero123 weights in csv config, and also use more view data to try, but I also can't get good result, so I want to ask for some advices

polinaverzun commented 5 months ago

same, the example car images in data folder give the same terrible result, my own photos of toy car (left, right, front, back) also make these weird artifacts

https://github.com/ashawkey/stable-dreamfusion/assets/156309711/e3435133-665f-46f6-be70-0bcebe6b367a

polinaverzun commented 5 months ago

https://github.com/ashawkey/stable-dreamfusion/assets/156309711/6f7d4523-62c7-43dc-9e31-9a1cc4262fc7