Commit Graph

83 Commits

Author SHA1 Message Date
63f373e7ec Adjust to updated skelda camera format. 2025-03-06 16:04:46 +01:00
651b806a3a Small updates. 2025-02-11 11:26:58 +01:00
2c994eca44 Allow keypoint detections outside the image. 2025-02-11 11:26:33 +01:00
c1cc6054d1 Some small updates. 2025-02-05 11:58:01 +01:00
b286efeb86 Updating jetson installation parts. 2025-02-05 10:33:02 +01:00
3eb7696e31 Some small updates. 2025-02-03 17:19:02 +01:00
b41380b033 Eval with all panoptic cameras. 2025-01-28 15:30:23 +01:00
321aa0a714 Increased 2D pose merging threshold. 2025-01-28 14:54:01 +01:00
1cc71cce57 Dropping more old python code. 2025-01-28 11:50:50 +01:00
56718838f2 Improved bad results debugging. 2025-01-27 16:49:47 +01:00
4b0d3a545e Updated results. 2025-01-27 16:47:34 +01:00
d8beaecc56 Changed detector for wb poses. 2025-01-27 15:46:55 +01:00
f9127e9a40 Merge close 2D poses. 2025-01-27 14:20:52 +01:00
d886d1db7c Fixed time calculation. 2025-01-27 14:18:08 +01:00
a1f85a5a53 Updated results. 2025-01-27 12:13:37 +01:00
9778c75bf0 Drop old code. 2025-01-21 18:01:14 +01:00
29c072400f Mixed updates and improvements. 2025-01-21 16:23:19 +01:00
c5f190ab35 Eval skelda datasets with cpp implementation. 2025-01-21 15:10:43 +01:00
d77fee7103 Restructuring some code. 2025-01-20 18:00:37 +01:00
8f2322694a Merge model outputs directly in graph. 2025-01-17 18:33:19 +01:00
8a249a2f16 Small improvements and fixes. 2025-01-17 16:24:37 +01:00
2f52521b9a Using bayer encoding for images. 2025-01-13 16:50:08 +01:00
b483269697 Renamed new wrapper script. 2024-12-19 14:23:41 +01:00
7b073249d6 Deleted old 2D pose wrapper. 2024-12-19 14:22:48 +01:00
5bc7bd9094 Splitting 2D and 3D predictions into single loops to prevent CPU priority issues. 2024-12-19 14:13:12 +01:00
2ad20549f2 Small updates. 2024-12-19 13:52:44 +01:00
19bb1a6372 Use fp32 model for whole-body poses. 2024-12-19 11:12:13 +01:00
1bb7dd966e Converted wb model as well. 2024-12-18 16:53:32 +01:00
f8984f9408 Some mixed updates. 2024-12-18 16:28:13 +01:00
07426fac2f Optional batched pose processing. 2024-12-18 16:22:08 +01:00
7b8d209601 Cache onnx-trt models, they are faster than using two trt-engines. Removed trt-runtime again. 2024-12-18 14:28:25 +01:00
b26ec998b3 Using tensorrt runtime directly. 2024-12-18 12:10:45 +01:00
21e93661f5 Updated dataset configs and results. 2024-12-16 14:43:54 +01:00
17ff41010a Some small updates. 2024-12-16 14:34:44 +01:00
07f75c53ee Renamed a parameter. 2024-12-09 14:17:35 +01:00
23108dd594 Further small improvements. 2024-12-06 18:15:08 +01:00
ee8b9bafb3 Updated easypose scripts. 2024-12-06 17:35:49 +01:00
108937d96c Some mixed improvements. 2024-12-06 13:58:34 +01:00
7a253cd615 Moved image cropping out of the graph again. 2024-12-06 12:32:51 +01:00
6452d20ec8 Some small updates. 2024-12-04 11:46:06 +01:00
97ff32b9ce Moved pose pre/post-processing into onnx graph. 2024-12-03 17:59:31 +01:00
742d2386c7 Moved detector pre/post-processing into onnx graph. 2024-12-03 11:52:55 +01:00
36781e616b Use rgb input for both models. 2024-12-02 17:28:34 +01:00
dc44a71b2c Some further speedups. 2024-12-02 17:00:04 +01:00
c8c48e4bf2 Using TensorRT backend. 2024-12-02 13:50:35 +01:00
9cc63532d3 Extracted onnx tools into a single script. 2024-12-02 12:32:53 +01:00
19015c6326 Exporting fp16 onnx models. 2024-11-29 17:37:44 +01:00
7b9505ca02 Moved image normalization into onnx models. 2024-11-29 17:05:43 +01:00
93d4611a91 Fixed running custom onnx models. 2024-11-29 15:18:57 +01:00
f6d13ea5a7 Testing onnx runtime with easypose. 2024-11-29 14:25:57 +01:00