Macbook m1 config for CPU ONLY processing

[itscharanteja]

./scripts/inference.sh ${CONFIG} ${CHECKPOINT} ${VIDEO_NAME} ${OUTPUT_DIR}, --pose_track
CONFIG - configs/halpe_26/resnet/256x192_res50_lr1e-3_1x.yaml (For Example)
CHECKPOINT - pretrained_models/halpe26_fast_res50_256x192.pth (You should download from MODEL_ZOO and place it in the "pretrained_models/")
VIDEO_NAME - /Users/charan/Downloads/videoplayback.mp4 (input video path)
OUTPUT_DIR - output path

This is the command to run and detect 2D keypoints.

[geokal]

Hi, I made a working version that uses PyTorch MPS (Apple Metal Performance Shaders) for acceleration) wherever it's possible. I used "configs/coco/resnet/256x192_res50_lr1e-3_1x.yaml" and "pretrained_models/fast_res50_256x192.pth". you just pass an extra "--device mps". video used video.mp4. Full command python scripts/demo_inference.py --cfg configs/coco/resnet/256x192_res50_lr1e-3_1x.yaml --checkpoint pretrained_models/fast_res50_256x192.pth --video /Users/georgek/Downloads/video_resized.mp4 --sp --device mps --detector yolo --outdir examples/res/ --save_video --vis_fast --debug --profile. it was run in MacBook Air m1 16gb and bumped to 30w total power used cpu nearly 90% and gpu 100% most of the time