Jiacong Fang

@Namburger Thanks for your suggestion. After checking the PyTorch and TF documentation, I found that PReLU is actually identical to LeakyReLU. I could just substitute LeakyReLU with PReLU without any...

@Namburger I find a way to suppress HardSwish transformation in TFLite conversion. Just comment out TensorFlow v.2.4.0 code: https://github.com/tensorflow/tensorflow/blob/582c8d236cb079023657287c318ff26adb239002/tensorflow/compiler/mlir/lite/transforms/optimize_patterns.td#L247-L286 and build TensorFlow from source. Then, one can use this modified...

@glenn-jocher @pirazor According to the updated YOLOv5 v4.0 release, I could convert all activation layers to Edge TPU. Inference time reduces from ~100ms (v3) to ~60ms (v4) for 320x320 input...

Plz check your CUDA version and your GPU model type. Or try running `models/tf.py` again after rebooting your machine? Actually, I didn't build TF with CUDA support because `models/tf.py` does...

@ling9601 Thanks. My colleague run YOLOv5 v4 320x320 on EdgeTPU for 24 hours yesterday without errors. He uses dpkg `libedgetpu1-max` library. We just plug Coral EdgeTPU to a USB dock...

@ling9601 I agree with you, but I can't assure that since I haven't an RPI.

@pliablepixels You could export and test Edge TPU model with https://github.com/zldrobit/yolov5/tree/tf-android-tfl-detect. PS: I attached the COCO Edge TPU model with 320x320 input resolution in case you're interested. [yolov5s-int8_edgetpu.zip](https://github.com/google-coral/edgetpu/files/5877537/yolov5s-int8_edgetpu.zip)

@pliablepixels plz report Edge TPU YOLOv5 related issues to https://github.com/ultralytics/yolov5/pull/1127

@batrlatom I use an Intel CPU with 20 threads for the test, and it should be faster than the CPU on the dev board. EDIT: There are still 12 operators...

It seems that you didn't load any pretrained weights.