refine: clarify backend install choices and provider detection

Agent-Logs-Url: https://github.com/IBB666/IBB_POSE/sessions/4a0abc4a-e2ce-442d-9e6f-7710a104d736

Co-authored-by: kukulu00 <1170651+kukulu00@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot]
2026-04-22 06:09:40 +00:00
committed by GitHub
co-authored by kukulu00
parent 3851a606d1
commit 0dd2dab1d2
2 changed files with 10 additions and 4 deletions
+3 -1
View File
@@ -33,6 +33,8 @@ Inspired by [SDPose-OOD](https://github.com/judian17/ComfyUI-SDPose-OOD).
```
3. Install only the backend you plan to use:
```bash
# Choose ONE backend path based on your workflow / hardware:
# Body / OpenPose
pip install ultralytics
@@ -48,7 +50,7 @@ Inspired by [SDPose-OOD](https://github.com/judian17/ComfyUI-SDPose-OOD).
### Windows / ComfyUI install note
- `IBB_POSE` no longer requires `groundingdino-py`, `chumpy` or `huggingface_hub` during node installation.
- If your embedded ComfyUI Python had previous failed installs cached, remove those partial packages before retrying, for example:
- If your embedded ComfyUI Python had previous failed installs cached, remove those partial packages before retrying, for example. Only do this if you know other nodes do not rely on those packages:
```bash
python -m pip uninstall -y groundingdino-py chumpy huggingface_hub
python -m pip cache purge
+7 -3
View File
@@ -723,11 +723,15 @@ class IBBLoadPoseModel:
if dev.type != "cpu":
cuda_prov = ["CUDAExecutionProvider", "CPUExecutionProvider"]
try:
_t = ort.InferenceSession(pose_path, providers=cuda_prov)
providers = cuda_prov
del _t
available = set(ort.get_available_providers())
if "CUDAExecutionProvider" in available:
providers = cuda_prov
else:
raise RuntimeError("CUDAExecutionProvider unavailable")
except Exception:
print("IBB_POSE: ONNX CUDAExecutionProvider unavailable – using CPU.")
else:
providers = cuda_prov
print(f"IBB_POSE: Loading DWPose ONNX | providers: {providers}")
det_session = ort.InferenceSession(det_path, providers=providers)