Note: Unfortunately, this will not work on CPU given the huge ammount of memory required. Your only option with CPU-only versions is to manually crop the people to fit the whole area of the image that is fed into OpenPose.
This command provides the most accurate results we have been able to achieve for body, hand and face keypoint detection.
This command provides the most accurate results we have been able to achieve for body, hand and face keypoint detection. However, this command will need ~10.5 GB of GPU memory (6.7 GB for COCO model) and runs at ~2 FPS on a Titan X for the body-foot model (1 FPS for COCO).
However:
- This will not work on CPU given the huge ammount of memory required. Your only option with CPU-only versions is to manually crop the people to fit the whole area of the image that is fed into OpenPose.
-**Note 1:** Increasing `--net_resolution` will highly reduce the frame rate and increase latency, while it might increase the accuracy. However, this accuracy increase is not guaranteed in all scenarios, required a more detailed analysis for each particular scenario. E.g., it will work better for images with very small people, but usually worse for people taking a big ratio of the image. Thus, we recommend to follow the commands below for maximum accuracy in most cases for both big and small-size people.
- It will also need ~10.5 GB of GPU memory for body-foot (BODY_25) model (~6.7 GB for COCO model).
-**Note 2: Do not use this configuration for MPII model**, its accuracy might be harmed by this multi-scale setting. This configuration is optimal only for COCO and COCO-extended (e.g., the default BODY_25) models.
- This requires GPUs like Titan X, Titan XP, some Quadro models, P100, V100, etc.
- Including hands and face will require >= 16GB GPUs (so the 12 GB GPUs like Titan X and XPs will no longer work).
- This command runs at ~2 FPS on a Titan X for the body-foot model (~1 FPS for COCO).
- Increasing `--net_resolution` will highly reduce the frame rate and increase latency, while it might increase the accuracy. However, this accuracy increase is not guaranteed in all scenarios, required a more detailed analysis for each particular scenario. E.g., it will work better for images with very small people, but usually worse for people taking a big ratio of the image. Thus, we recommend to follow the commands below for maximum accuracy in most cases for both big and small-size people.
-**Do not use this configuration for MPII model**, its accuracy might be harmed by this multi-scale setting. This configuration is optimal only for COCO and COCO-extended (e.g., the default BODY_25) models.
