search

hunglc007 / tensorflow-yolov4-tflite

YOLOv4, YOLOv4-tiny, YOLOv3, YOLOv3-tiny Implemented in Tensorflow 2.0, Android. Convert YOLO v4 .weights tensorflow, tensorrt and tflite
https://github.com/hunglc007/tensorflow-yolov4-tflite
MIT License
2.25k stars 1.24k forks source link

detection box too small #159

Open Eve66666 opened 4 years ago

Eve66666 commented 4 years ago

When I used your source code to transform the yolov4 model I had trained, the TF weight file I got was tested. The detection box shown in the test results was too small to completely cover the whole target. What's the matter?

image

mhcao916 commented 4 years ago

hi, this project can be used directly to transfer learning? when i run it on my own dataset, i get a trained file, but after i convert it, the model doesn't work, why? even i set my class num to 80, in other word, i train it on coco dataset, and i convert my train model, it failed to detect object, why? i use tf-gpu 2.3rc0, cuda 9.1, it is because cuda version is too low to work well?

Eve66666 commented 4 years ago

Sorry, I can't solve this. I trained my own model through DarkNet and then converted to. Pb file through the code here

Gooogr commented 4 years ago

Check this issue: https://github.com/hunglc007/tensorflow-yolov4-tflite/issues/174 The problem was in default anchors values in the core/config.py