Remove hard-coded batch size of 1 from YOLO Exporter #932
Conversation
This line prevents users from providing a custom batch size to the exporter with kwargs in `model.export`. When it's removed, providing a batch size works as expected. I verified this by exporting one of my v8n models to onnx and providing a custom batch size, then opening the model on netron.app.
```
from ultralytics import YOLO
model = YOLO("trained_models/v8n.pt")
model.export(format="onnx", batch=8)
```
It looks like this is an artifact from a work-around introduced in ultralytics#129. In ultralytics#132 the conditional was commented out but the body was left in.
|
CLA Assistant Lite bot All Contributors have signed the CLA. ✅ |
|
I have read the CLA Document and I sign the CLA |
|
@EricPedley thanks for the PR! Unfortunately this placeholder helps exports default to batch=1, otherwise they will use the default.yaml batch=16, which is probably not the dominant use case here. We need to figure out a better way to set batch size per task and make this more flexible so that i.e. exports default to batch=1, train defaults to batch=16 etc., while still allowing overrides when batch argument is passed. |
|
@EricPedley resolved in conjunction with 8.0.35 PR to allow setting default 1 but also allow manually defined batch sizes. Only problem is if they match default.yaml batch=16 then it will get reset to 1. All others are allowed, i.e. |
2 participants
This line prevents users from providing a custom batch size to the exporter with kwargs in
model.export. When it's removed, providing a batch size works as expected. I verified this by exporting one of my v8n models to onnx and providing a custom batch size, then opening the model on netron.app.It looks like this is an artifact from a work-around introduced in #129. In #132 the conditional was commented out but the body was left in. (I figured this out with GitLens. It's really cool and worth checking out)
🛠️ PR Summary
Made with ❤️ by Ultralytics Actions
🌟 Summary
This Pull Request has no code changes.
📊 Key Changes
🎯 Purpose & Impact