DeepSparkHub/3d-reconstruction/ngp-nerf/pytorch

• HashNeRF
• • Model description
  • Step 1: Installation
  • Step 2: Preparing datasets
  • Step 3: Training and test
  • • One single GPU
  • Results on BI-V100
  • Reference

HashNeRF

Model description

A PyTorch implementation (Hash) of the NeRF part (grid encoder, density grid ray sampler) in instant-ngp, as described in Instant Neural Graphics Primitives with a Multiresolution Hash Encoding.

Step 1: Installation

pip3 install -r requirements.txt

Step 2: Preparing datasets

We use the same data format as instant-ngp, fox and blender dataset nerf_synthetic.Please download and put them under ./data.

For custom dataset, you should:

1. take a video / many photos from different views
2. put the video under a path like ./data/custom/video.mp4 or the images under ./data/custom/images/*.jpg.
3. call the preprocess code: (should install ffmpeg and colmap first! refer to the file for more options)

python3 scripts/colmap2nerf.py --video ./data/custom/video.mp4 --run_colmap # if use video
python3 scripts/colmap2nerf.py --images ./data/custom/images/ --run_colmap # if use images

Step 3: Training and test

One single GPU

First time running will take some time to compile the CUDA extensions.

# train with fox dataset
python3 main_nerf.py data/fox --workspace trial_nerf -O
# data/fox is dataset path; --workspace means output path; -O means --fp16 --cuda_ray --preload, which usually gives the best results balanced on speed & performance.

# test mode
python3 main_nerf.py data/fox --workspace trial_nerf -O --test

# train with the blender dataset, you should add `--bound 1.0 --scale 0.8 --dt_gamma 0`
# --bound means the scene is assumed to be inside box[-bound, bound]
# --scale adjusts the camera locaction to make sure it falls inside the above bounding box. 
# --dt_gamma controls the adaptive ray marching speed, set to 0 turns it off.
python3 main_nerf.py data/nerf_synthetic/lego --workspace trial_nerf -O --bound 1.0 --scale 0.8 --dt_gamma 0

# train with custom dataset(you'll need to tune the scale & bound if necessary):
python3 main_nerf.py data/custom_data --workspace trial_nerf -O

Results on BI-V100

@@ -65,90 +60,4 @@ python3 main_nerf.py data/nerf_synthetic/lego --workspace trial_nerf -O

Convergence criteria	Configuration (x denotes number of GPUs)	Performance	Accuracy	Power（W）	Scalability	Memory utilization（G）	Stability
0.0652	SDK V2.2,bs:1,1x,fp16	10	11.9	82	0.903	28.1	1

Reference

• torch-ngp
• DearPyGui