Downloads · 30 days
0
depthsplat
Yeqing0814/depthsplat
depthsplat is a machine learning model from Yeqing0814. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
- 因为用到了稀疏3D卷积，需要安装MinkowskiEngine - 但是对于cuda版本大于11的安装这个库存在很多环境问题： - 按照以下步骤可以解决 - （1）首先按照这个链接安装环境(其中Step 4与Step 5已完成)： https://github.com/Julie-tang00/Common-envs-issues/blob/main/Cuda12-MinkowskiEngine
Access
Public
Updated Sep 13, 2025
Repo size
6.6 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.ckpt6.6 GB · 99%
At a glance
- Access
- Public
- Created
- Sep 10, 2025
- Updated
- Sep 13, 2025
- SHA
- a6dd0409
Model card
From the Hugging Face model README
安装MinkowskiEngine
-
因为用到了稀疏3D卷积，需要安装MinkowskiEngine
-
但是对于cuda版本大于11的安装这个库存在很多环境问题：
-
按照以下步骤可以解决
-
（1）首先按照这个链接安装环境(其中Step 4与Step 5已完成)： https://github.com/Julie-tang00/Common-envs-issues/blob/main/Cuda12-MinkowskiEngine
-
（2）安装依赖：
conda activate depthsplat
conda install -c conda-forge openblas
- （3）测试：
python
>>> import MinkowskiEngine as ME
>>> print(ME.__version__)
0.5.4
训练指令
conda activate depthsplat
CUDA_VISIBLE_DEVICES=2 python -m src.main +experiment=dl3dv \
data_loader.train.batch_size=1 \
'dataset.roots'='["datasets/dl3dv"]' \
dataset.view_sampler.num_target_views=8 \
dataset.view_sampler.num_context_views=6 \
dataset.min_views=1 \
dataset.max_views=6 \
trainer.max_steps=100000 \
trainer.num_nodes=1 \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.monodepth_vit_type=vitb \
checkpointing.pretrained_monodepth=pretrained/pretrained_weights/depth_anything_v2_vitb.pth \
checkpointing.pretrained_mvdepth=pretrained/pretrained_weights/gmflow-scale1-things-e9887eda.pth \
wandb.project=VolSplat \
output_dir=checkpoints/dl3dv-256x448-depthsplat-base-randview2-6 \
train.no_viz_video=True
- Created
- Sep 10, 2025
- Updated
- Sep 13, 2025