Glenn Jocher

@chmjelek it's possible certain EXIF info may not be read correctly from all formats, especially for manually rotated videos. We use two 4k iPhone-captured videos to verify video inference, one...

@DoAlone we use two reference iPhone videos for testing, both work correctly for us. ``` python detect.py --source https://github.com/ultralytics/yolov5/releases/download/v1.0/decelera_landscape.mov python detect.py --source https://github.com/ultralytics/yolov5/releases/download/v1.0/decelera_portrait.mov ```

@DoAlone if these work correctly for you then please try to provide us a reproducible workflow to create your inverted videos on our side. If we can reproduce we can...

@DoAlone I'm able to reproduce this now on iPhone 12 videos (4k 30 FPS) recorded today and airdropped to my Macbook in *.mov format. It seems that cv2.CAP_PROP_ORIENTATION_AUTO is failing...

@DoAlone since this seems like a flaw in the cv2 video capure code I'd comment directly on https://github.com/opencv/opencv/issues/15499 It doesn't make sense for us to patch a solution upstream when...

@DoAlone got it. Since it seems like a cv2 bug it's probably worth raising issue on the cv2 repo.

@PraveenMNaik 👋 Hello! Thanks for asking about YOLOv5 🚀 **architecture visualization**. Fusing Conv+BatchNorm layers will reduce layer and parameter counts. We've made visualizing YOLOv5 🚀 architectures super easy. There are...

@Haoyanlong very interesting! Do you have the full benchmarking script? I wonder if you need to make the images and model parameters contiguous after the change? https://pytorch.org/docs/stable/generated/torch.Tensor.contiguous.html

@tai-nguyenh-tpv-otsv in python you can delete any variable you want with del: ``` del model ```

@PhantomHydrino 👋 Hello! Thanks for asking about **handling inference results**. YOLOv5 🚀 [PyTorch Hub](https://docs.ultralytics.com/yolov5/tutorials/pytorch_hub_model_loading) models allow for simple model loading and inference in a pure python environment without using `detect.py`....