Manydepth 环境配置及测试

1. 源码下载

paper：https://arxiv.org/abs/2104.14540
github：https://github.com/nianticlabs/manydepth

git clone https://github.com/nianticlabs/manydepth.git

2. 环境配置

参看environment.yml，需要安装以下库

name: manydepth
channels:
  - default
  - pytorch
  - conda-forge
dependencies:
  - opencv=3.4.2
  - matplotlib=3.1.2
  - numpy=1.19.5
  - pip
  - pip:
    - Pillow==6.2.1
    - tensorboardX==1.5
    - scikit-image==0.16.2
    - torch==1.7.1
    - torchvision==0.8.2
    - tqdm==4.57.0

安装操作：

conda create -n manydepth python=3.8
conda activate manydepth 

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple matplotlib numpy Pillow tensorboardX scikit-image tqdm opencv-python

3. 测试

下载模型

You can download weights for some pretrained models here:

• KITTI MR (640x192)
• KITTI HR (1024x320)
• CityScapes (512x192)

放置在weight 目录下

测试

python -m manydepth.test_simple \
    --target_image_path assets/test_sequence_target.jpg \
    --source_image_path assets/test_sequence_source.jpg \
    --intrinsics_json_path assets/test_sequence_intrinsics.json \
    --model_path weight/KITTI_MR/