windingwind/seal-3d

The official implementation of the paper Seal-3D: Interactive Pixel-Level Editing for Neural Radiance Fields, the first interactive pixel-level NeRF editing tool.

seal-3d's Issues

Areas not selected by the brush are edited

Brush editing commands 1：

Edit results：
view 1

view 2

view 3

Brush editing commands 2：

Edit results：
view 1

view 2

view 3

I'm wondering why the large area in the middle of the right wheel is also edited. Hope you can give some advice.

About Bounding Box tool confusion

2D Bounding Box How to select a 3D target

non-rigid blending support

Looking at section 4.2 in your excellent paper, I have noticed you mentioned a way for doing non-rigid blending, which I find very interesting. Are you planning to release support for this feature as well?

CUDA out of memory

If I want to deploy the model on a low-end graphics card, what parameters do I need to change to reduce the level of memory required

seal_config for custom data

Hi,

Thanks for the great work!
I'm trying to use this for my custom capture. May I ask how shall I obtain the seal_config file for my custom data? Thanks!

headless mode fail, But gui mode works fine

When I don't use gui mode, it seems to fail during the test phase
CUDA11.6、gcc10.3、torch1.12.1 ubuntu18.04

Loading model from: /home/user/.conda/envs/torch-ngp/lib/python3.8/site-packages/lpips/weights/v0.1/alex.pth
[INFO] Trainer: ngp | 2023-08-29_11-03-50 | cuda | fp16 | exps/lego_ngp
[INFO] #parameters: 24490848
[INFO] Loading exps/lego_ngp/checkpoints/ngp_ep0150.pth ...
[INFO] loaded model.
[INFO] load at epoch 150, global step 30000
[INFO] loaded optimizer.
[INFO] loaded scheduler.
[INFO] loaded scaler.
[INFO] Trainer: ngp | 2023-08-29_11-03-50 | cuda | fp16 | exps/lego_brush_compare
[INFO] #parameters: 24490848
[INFO] Loading exps/lego_ngp/checkpoints/ngp_ep0150.pth ...
[INFO] loaded model.
[INFO] load at epoch 150, global step 30000
[INFO] loaded optimizer.
[INFO] loaded scheduler.
[INFO] loaded scaler.
Local x generation: 0.873894214630127
Surrounding x generation: 1.1920928955078125e-06
Global x generation: 4.76837158203125e-07
Loading train data: 100%|█████████████████████████████████████████| 100/100 [00:01<00:00, 59.49it/s]
Loading val data: 100%|███████████████████████████████████████████| 100/100 [00:01<00:00, 59.85it/s]
[INFO] Proxy train/eval/test: True/True/False
Proxying train data: 0%| | 0/100 [00:00<?, ?it/s]段错误 (核心已转储)

Instant NGP false

I train the Instant NGP model with 360_v2 dataset， but the PSNR is low to 10。Can you afford 1 or 2 pretrained model with your dataset ？Thank you very much！email : [email protected]