（深度估计学习）Depth Anything V2 复现

代码：https://github.com/DepthAnything/Depth-Anything-V2

一、配置环境

在本机电脑win跑之后依旧爆显存，放到服务器跑：Ubuntu22.04，CUDA17

conda create -n DAv2 python=3.10
conda activate DAv2

conda下安装cuda。由于服务器上面我不能安装CUDA，只能在conda上安装cuda。我安装的cuda11.7。
跟着下面的教程做：

conda虚拟环境中安装cuda和cudnn，再也不用头疼版本号的问题了

wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/linux-64/cudatoolkit-11.7.1-h4bc3d14_13.conda
conda install --use-local cudatoolkit-11.7.1-h4bc3d14_13.conda
wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/linux-64/cudnn-8.9.7.29-hcdd5f01_2.conda
conda install --use-local cudnn-8.9.7.29-hcdd5f01_2.conda

安装其他依赖
记得在requirements.txt中增加tensorboard、h5py

pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

检查torch是否安装正确以及cuda版本

python
import torch
torch.cuda.is_available()
torch.version.cuda

二、准备数据

1. 权重文件

将pre-trained-models放在 DepthAnythingV2/checkpoints 文件夹

2. 训练数据

训练的时候需要，我这里之前就准备了vkitti。我先用vkitti数据跑一下试一下。

三、Test

Running script on images：

python run.py \--encoder <vits | vitb | vitl | vitg> \--img-path <path> --outdir <outdir> \[--input-size <size>] [--pred-only] [--grayscale]

Options:

• –img-path: You can either 1) point it to an image directory storing all interested images, 2) point it to a single image, or 3)
  point it a text file storing all image paths.
• –input-size (optional): By default, we use input size 518 for model inference. You can increase the size for even more fine-grained
  results.
• –pred-only (optional): Only save the predicted depth map, without raw image.
• –grayscale (optional): Save the grayscale depth map, without applying color palette.

For example:

python run.py --encoder vitl --img-path assets/examples --outdir depth_vis

Running script on videos

python run_video.py \--encoder <vits | vitb | vitl | vitg> \--video-path assets/examples_video --outdir video_depth_vis \[--input-size <size>] [--pred-only] [--grayscale]

Our larger model has better temporal consistency on videos.

四、Train

根据自己的数据修改DepthAnythingV2/metric_depth/dataset/splits和train.py中的路径数据

sh dist_train.sh

但我运行不了这个sh文件，所以我选择直接配置.vscode/launch.json。并且我将我的train代码改为了非分布式的。

{// 使用 IntelliSense 了解相关属性。 // 悬停以查看现有属性的描述。// 欲了解更多信息，请访问: https://go.microsoft.com/fwlink/?linkid=830387"version": "0.2.0","configurations": [{"name": "Python 调试程序: train.py","type": "debugpy","request": "launch","program": "${workspaceFolder}/metric_depth/train.py","console": "integratedTerminal","args": ["--epoch", "120","--encoder", "vitl","--bs", "2","--lr", "0.000005","--save-path", "./exp/vkitti","--dataset", "vkitti","--img-size", "518","--min-depth", "0.001","--max-depth", "20","--pretrained-from", "./checkpoints/depth_anything_v2_vitl.pth", ],"env": {"MASTER_ADDR": "localhost","MASTER_PORT": "20596"}},{"name":"Python 调试程序: run.py","type": "debugpy","request": "launch","program": "${workspaceFolder}/run.py","console": "integratedTerminal","args": ["--encoder", "vitl","--img-path", "assets/examples","--outdir", "output/depth_anything_v2_vitl_test","--checkpoints","checkpoints/depth_anything_v2_vitl_test.pth"],}]
}