wenjinf0811 commited on
Commit
f2c7ced
·
verified ·
1 Parent(s): 744c163

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +26 -0
README.md ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Person Following — Model Files
2
+
3
+ Models consumed by `person_following.gesture` and the YOLO detection
4
+ pipeline. Mix of:
5
+
6
+ * **ONNX sources** — portable, compile locally to TRT on the target device.
7
+ * **TRT engines** — GPU-specific prebuilt for faster cold start.
8
+ Falls back to ONNX-compile if the engine fails compatibility check.
9
+ * **PyTorch checkpoints** — loaded directly by Python code.
10
+
11
+ ## Files
12
+
13
+ | Name | Type | Notes |
14
+ |---|---|---|
15
+ | `rtmw.onnx` | ONNX | RTMW pose, 133 keypoints, 384x288 |
16
+ | `rtmw_l_fp16_dynbatch.engine` | TRT engine | FP16, dynamic batch 1-6, built on Thor |
17
+ | `action_recog_stgcn.pth` | PyTorch | 7-class HRI gesture (OccAug v3) |
18
+ | `yolo11n.engine` | TRT engine | Person detection (built on Thor) |
19
+ | `yolo11s-seg.engine` | TRT engine | Person segmentation (built on Thor) |
20
+
21
+ ## Engine portability caveat
22
+
23
+ TRT engines are not portable across GPU architectures or TRT versions.
24
+ The engines here were built for **NVIDIA Jetson Thor / TRT 10.16**. On
25
+ other hardware, `ModelManager` will detect the incompatibility, discard
26
+ the downloaded engine, and rebuild from ONNX.