Zero-1-to-3 Reproduction Log

Zero-1-to-3 (based on threestudio)

Zero-1-to-3: Zero-shot One Image to 3D Object
Liu et al. | CVPR 2023

---

Paper Summary

---

Installation

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install ninja
pip install -r requirements.txt

---

Pre-trained Weights

cd load/zero123
wget https://huggingface.co/cvlab/zero123-weights/resolve/main/105000.ckpt

---

Reconstruction

• 'zero123.yaml' 파일 중 image_path $\rarr$ ./load/images/000.png 수정

python launch.py --config configs/zero123.yaml --train --gpu 0

---

7/6 Thu.

• ./load/images/hamburger.png 사용하여 25,000-it 학습.
• 1,000 steps 이후로는 학습이 정체되더니 17,000 steps에서 모델링이 무너져 사라짐.
  *영상은 1) RGB, 2) Normal, 3) Texture, 4) Depth 순서

[Reference: 25,000-it]

---

*Server: V-100 32GB
[50-it | 1min 4sec]

[100-it | 2min 28sec]

[200-it | 4min 17sec]

[500-it | 10min 44sec]

[1,000-it | 21min 28sec]

[15,000-it | 5hr 22min]

[17,050-it | 6hr 6min]

[17,250-it | 6hr 10min]

[25,000-it | 8hr 57min]

---

7/11 Mon.

[500-it | 10min]