Nicklas Börjesson
Nicklas Börjesson
Hi! Ok, so that was faster, 29 seconds, about 6 seconds faster. However, the boxes change a bit, the confidence seem to lower slightly with the patch. It has a...
WRT RELU, I'll have to retrain it then, afaik it is a bit cumbersome to change from the defaults in Yolo, I'll be back with that next week, have a...
Annoyingly, something has broken in my Yolov5n export and int8 quantization which breaks the model in some way I have yet to be able to fix, I can't successfully export...
> Hi @nicklasb I used an existing model at GH user already exported from [here](https://github.com/muhammedakyuzlu/yolov5-tflite-cpp) which fits our requirement. > Thanks. It seems exactly like what I have. I think...
@vikramdattu Ok, so now I have retrained with act: nn.ReLU() in the yaml settings. I *hope* that it actually used ReLU, I couldn't see any difference in the logging output,...
> Cannot be charged via 5V Pin. Does that mean that the only way to charge it is using the USB? Is there no other way but adding an external...
> When T-Beam v1.1 was first designed, there was no external power supply input interface reserved other than USB. Does this indicate that there are less hacky methods for the...
Well, yeah, something like that would be the usage, I suppose. As close as the normal ASF usage as possible. Have you looked closer to see if it easy to...
Took a quick look, and the code seems really nice, it should be pretty easy to break that out.
Please do, it would be a killer feature if you could get that done.