Release: MediaPipe segfault fix, RTMPose optimization, UI improvements#80
Merged
Conversation
… box colors Mirror camera frames horizontally (selfie-style) before sending to the backend so the streaming video matches what the user sees. Only applies to camera source, not video file uploads. Also darken the object detection bounding box color palette so white label text is readable on screen. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: mirror camera feed and darken bounding box colors
The SSH user does not have write permission to /var/www. Use sudo mkdir and chown to match the ownership pattern of the production directory. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This reverts commit 3ba82ae.
Fix/small bugs
…O mode Eliminates segfaults in libmediapipe.so caused by async detect_async() race conditions in the native C++ layer (google-ai-edge/mediapipe#6085, #5466). - mediapipe_processor.py: remove ObjectDetector (now separate processor), switch PoseLandmarker to synchronous detect_for_video() - mediapipe_object_detector_processor.py: switch to detect_for_video() - mediapipe_hand_gesture_processor.py: switch to recognize_for_video() - View2D: lock container to 4:3 aspect ratio for consistent sizing - Skeleton3DViewer: zoom out camera to show full avatar + video Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: MediaPipe segfault, camera mirror, UI improvements
Lock container to 4:3 aspect ratio with margin auto centering. Remove objectFit: contain which caused variable black bar sizing across functions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…scale Apply fixed 0.25 scale factor to RTMPose world landmark coordinates so the 3D skeleton matches MediaPipe's normalized coordinate space. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: RTMPose coordinate normalization, View2D sizing
perf: RTMPose detection frame skipping (det_frequency=7)
Change hand gesture colors from bright cyan/magenta to dark teal/magenta so white label text is readable against the bounding box backgrounds. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…colors Add step to uninstall CPU onnxruntime and reinstall onnxruntime-gpu after pip install requirements.txt, which pulls in the CPU variant via rtmlib dependency. Fixes RTMPose running on CPU instead of GPU after deploys. Also darken hand gesture bounding box colors for readability. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: onnxruntime GPU in deploy pipelines, darken hand gesture colors
…U variant Uninstalling only the CPU onnxruntime removes shared module files that onnxruntime-gpu also uses, breaking the installation. Now uninstall both packages and force-reinstall onnxruntime-gpu to ensure a clean state. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: onnxruntime-gpu deploy pipeline broken installation
Replace perspective unprojection (which caused skeleton to shrink/grow with arm movement due to body_height_px dependency) with direct use of RTMPose3D's root-relative keypoints_3d. Perspective unprojection now only computes root_position for scene placement. Removes scale hack. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: stable RTMPose skeleton scale using root-relative 3D coords
Model keypoints_3d x,y are in model-input-space (~288x384 pixels) while z is already in meters. Apply _XY_SCALE (z_range/z_input_half ≈ 0.015) to convert x,y to meters, matching the z coordinate space. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: scale RTMPose skeleton x,y to meters
…inates Previously only z was decoded from simcc while x,y came from keypoints_3d which depends on the bbox affine crop — causing skeleton size to change when the detection bbox shifts. Now all three axes are decoded from the raw simcc codec space (input_size 288x384x288), making coordinates consistent regardless of bbox changes. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: bbox-independent RTMPose skeleton via full simcc decoding
…only Simcc x,y are bbox-crop-dependent — they scale when the detection bbox changes size. Instead, use keypoints_2d (stable image-space pixels) for x,y via perspective unprojection with a shared z_root for all joints. Skeleton proportions now come from stable 2D pixel ratios. Z depth still from corrected simcc (root-relative, bbox-independent). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fix: stable RTMPose skeleton using 2D pixels + simcc z
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Test plan
Generated with Claude Code