Skip to content

README.md

Latest commit

 

History

History
131 lines (96 loc) · 4.07 KB

README.md

File metadata and controls

131 lines (96 loc) · 4.07 KB

Tools Usage Guide

English | 中文

This directory contains tool scripts for the DepthAnything-AC project.

Inference Script (infer.py)

The inference script supports processing of both images and videos, with flexible batch processing capabilities.

Basic Usage

Single Image Inference

# Basic usage
python tools/infer.py --input image.jpg --output depth.png

# Specify model and encoder
python tools/infer.py --input image.jpg --output depth.png --model checkpoints/depth_anything_AC_vits.pth --encoder vits

# Use different colormap
python tools/infer.py --input image.jpg --output depth.png --colormap inferno

Single Video Inference

# Basic video processing
python tools/infer.py --input video.mp4 --output depth_video.mp4

# Specify output FPS
python tools/infer.py --input video.mp4 --output depth_video.mp4 --fps 30

# Use different colormap for video
python tools/infer.py --input video.mp4 --output depth_video.mp4 --colormap spectral

Batch Processing

# Process images only in a directory
python tools/infer.py --input images/ --output output/ --mode images

# Process videos only in a directory
python tools/infer.py --input videos/ --output output/ --mode videos

# Process both images and videos (default mode)
python tools/infer.py --input media/ --output output/ --mode mixed

# Recursive processing of subdirectories
python tools/infer.py --input dataset/ --output results/ --recursive

Parameters

Parameter Short Type Default Description
--input -i str - Input image/video path or directory (required)
--output -o str - Output path (file or directory) (required)
--model -m str checkpoints/depth_anything_v2_vits.pth Model weight path
--encoder - str vits Encoder type (vits, vitb, vitl)
--colormap - str spectral Colormap (inferno, spectral, gray)
--depth_cap - float 80 Maximum depth value for capping
--fps - float None Output video FPS (defaults to input FPS)
--recursive -r flag False Search recursively in subdirectories
--mode - str mixed Processing mode (images, videos, mixed)

Supported File Formats

Images: .jpg, .jpeg, .png, .bmp, .tiff, .tif

Videos: .mp4, .avi, .mov, .mkv, .flv, .wmv, .webm, .m4v

Usage Examples

Example 1: Process Single Image

python tools/infer.py -i test_image.jpg -o result.png

Output:

  • result.png - Colored depth map
  • result_raw.npy - Raw depth data

Example 2: Process Single Video

python tools/infer.py -i test_video.mp4 -o depth_video.mp4 --fps 30

Output:

  • depth_video.mp4 - Video with depth visualization

Example 3: Batch Process Images Only

python tools/infer.py -i photos/ -o depth_results/ --mode images --recursive

Recursively processes all images in the photos/ directory and subdirectories, generating corresponding depth maps in the depth_results/ directory.

Example 4: Batch Process Videos Only

python tools/infer.py -i videos/ -o depth_videos/ --mode videos --fps 25

Processes all videos in the videos/ directory, generating depth videos with 25 FPS output.

Example 5: Mixed Batch Processing

python tools/infer.py -i media/ -o output/ --mode mixed --colormap inferno

Processes both images and videos in the media/ directory, using inferno colormap for visualization.

Output Structure

When processing directories, the output maintains the same folder structure as the input:

  • Images: input_name_depth.png + input_name_depth_raw.npy
  • Videos: input_name_depth.mp4

Training Scripts

train.sh

Script for launching distributed training.

bash tools/train.sh [NUM_GPUS] [PORT]

val.sh

Script for launching distributed evaluation.

bash tools/val.sh [NUM_GPUS] [PORT] [DATASET]

For more detailed information, please refer to the main project README documentation.