Skip to content

Latest commit

 

History

History
50 lines (43 loc) · 2.62 KB

run.md

File metadata and controls

50 lines (43 loc) · 2.62 KB
Raw

AlphaPose run.sh Usage & Examples

We provide a script run.sh to ease your effort of running our code. Here, we first list the flags of this script and then give some examples.

Flags

  • --gpu: Which GPU(s) to use.
  • --batch: Batch size of the pose estimation network for each GPU card.
  • --indir: Directory of the input images. All the images in the directory will be processed. Input dir should be RELATIVE PATH to AlphaPose directory.
  • --list: A text file list for the input images, each line is the ASOLUTE PATH to the image. Not co-appear with --indir
  • --video: Read video and process the video frame by frame.
  • --outdir: Output directory to store the human detection and pose estimation results.
  • --mode: fast/normal/accurate. We recommend using the mode 'normal'. Their differences are listed below.
MODE multi-scale human detection 4 crop pose estimation accuracy speed
fast no no 70.6 0.9x
normal yes no 71.6 1x
accurate yes yes 72.3 4.9x
  • --vis: If turned-on, it will visualize the results and save them as images. If the input is video, it will save the output as video (but without audio).
  • --sep: If turned on, it will save the json file for each image/frame of the input. Default is false.
  • --dataset: Follow the keypoints definition as COCO or MPII dataset. Default is 'MPII', Alternative option is 'COCO'.
  • --format: The format of the saved results. By default, it will save the output in COCO-like format. An alternative option is 'cmu', which saves the results in the format of CMU-Pose. For more details, see format.md

Examples

  • Run AlphaPose for all images in a folder and display the results:
./run.sh --indir examples/demo/ --outdir examples/results/ --vis
  • Run AlphaPose for images on a list and save the results in COCO dataset's keypoints order:
./run.sh --list examples/img-list.txt --outdir examples/results/ --dataset COCO
  • Run AlphaPose for a video, save the results in CMU-Pose's format and display the results:
./run.sh --video examples/input.mp4 --outdir examples/results/ --vis --format cmu
  • Run AlphaPose for all images in 'fast' mode and save the result for each image seperately:
./run.sh --indir examples/demo/ --outdir examples/results/ --mode fast --sep
  • Speed up AlphaPose by using multi-gpu and larger batch size. Assumes that you have 2 GPU cards, each card has a memory of 8GB.:
./run.sh --indir examples/demo/ --outdir examples/results/ --gpu 0,1 --batch 6