Implementation of yolo-fastest and wrong prediction

[gitE0Z9]

Forked: https://github.com/gitE0Z9/tensorflow-yolov4-tflite

I follow the same pattern in this project to implement yolo-fastest-1.1(https://github.com/dog-qiuqiu/Yolo-Fastest/tree/master/ModelZoo/yolo-fastest-1.1_coco) and have finished most of work. However I got wrong prediction than what the model should produce.

Can you help me find out where the problem is? I have examined the model structure and weights, everything is fine, but the feature map is different from original model.

example code

git clone https://github.com/gitE0Z9/tensorflow-yolov4-tflite

cd tensorflow-yolov4-tflite

wget https://github.com/dog-qiuqiu/Yolo-Fastest/raw/master/ModelZoo/yolo-fastest-1.1_coco/yolo-fastest-1.1.weights -O fast.weights

python save_model.py --weights fast.weights --output fast --input_size 320x320 --model yolofastest --num_classes 80 --tiny

python detect.py --weights fast --size 320x320 --model yolofastest --image ./data/dog.jpg --tiny --output result.png