Model Won’t Load

Symptoms:

  • Clicking a model in the Models tab does nothing
  • Notice bar says “Failed to load model: …”
  • Aim Tools page stays in the “no model loaded” state

Checklist

1. Is the file actually an ONNX file?

PowerAim accepts only .onnx. Verify:

  • File extension is .onnx
  • File size is reasonable (small models 5–20 MB, large 50–250 MB)
  • File isn’t 0 bytes (a failed download)

If you downloaded the file from a non-store source, re-download it. Corrupted ONNX files refuse to load.

2. Is the file in the right folder?

PowerAim looks in <install dir>\bin\models\. If the file is anywhere else, it won’t appear in the local list.

3. Is the architecture supported?

PowerAim is built around YOLOv8 ONNX (anchor-free, NCHW input, 1×3×H×W). Other architectures may fail to load. Specifically:

  • YOLOv8 / YOLO11 detection heads ([1, 4+classes, anchors] or transposed) work
  • YOLOv5 ONNX works when the model declares its class names in its ONNX names metadata (current YOLOv5 exports do). That is how PowerAim recognises the extra objectness channel
  • End-to-end exports with NMS built in (YOLOv10, YOLO26, output [1, 300, 6]) work
  • Pose models (kpt_shape metadata) and segmentation models work; keypoints are read, masks are ignored
  • YOLOv7 should work but isn’t officially tested
  • Non-YOLO ONNX (custom architectures) likely fails

4. CUDA build trying to load on a non-NVIDIA system?

The CUDA build expects:

  • An NVIDIA GPU
  • The CUDA 12.x runtime installed on the system

If you have an AMD or Intel GPU and the CUDA build, PowerAim falls back to DirectML, but if CUDA itself crashes on load, the fallback never gets a chance. Switch to the DirectML build.

5. Test with a known-good model

Click any of the bundled Universal*.onnx files. They’re known to work. If even bundled models fail, the issue is environmental (.NET, VC++, ONNX Runtime) rather than the model.

6. Read the in-app log

Logs sidebar item. ONNX Runtime errors are written here verbatim, and they usually say exactly what’s wrong:

  • Unsupported operator: ... → model uses a custom op PowerAim doesn’t support
  • Invalid input shape: ... → input shape isn’t what PowerAim expects
  • DXGI device removed → GPU driver crashed; reboot

7. Re-install prerequisites

If models fail with cryptic native errors (access violation, DllNotFound), reinstall:

  • .NET Runtime 10 (x64)
  • Visual C++ 2015-2022 Redistributable (x64)

Both must be x64. PowerAim does not work with x86 variants.

8. Switch DirectML / CUDA build

Settings → UI Settings → Switch to DirectML / CUDA. If one build refuses to load a model, the other might handle it. CUDA in particular is more permissive about non-standard ops.

Multi-class model class list is empty

If the model loads but the Target Classes dialog shows Class 0, Class 1, etc. instead of names, the model is missing the names custom metadata field. Either:

  • Re-export from Ultralytics: model.export(format="onnx") automatically writes names
  • Or live with the numeric IDs; they still work for filtering

“Invalid input shape” specifically

PowerAim expects [N, 3, H, W] (NCHW) where N=1 and 3 is RGB channels. If your model is NHWC, re-export with NCHW order:

model.export(format="onnx", dynamic=True, simplify=True)

PowerAim is source-available under PolyForm Noncommercial. Commercial use prohibited.

This site uses Just the Docs, a documentation theme for Jekyll.