How is get_cls_scores implemented in yolov4_helper.py?

[tanhuakun]

If you have the time, could you explain how the class scores are calculated?

Is the output of darknet_model(img) a list of class confidence levels?

What do the scores represent? Do they represent a class confidence level for each class in a picture?

    def get_cls_scores(self, img:torch.tensor):
        img = _input_transform(img).to(device)
        output = self.darknet_model(img)  # <-- what does the output represent exactly?
        self.features = self.darknet_model.features
        scores = []
        for item in output:
            h, w = item.shape[-2], item.shape[-1]
            item = item.reshape(-1, 5+80, h*w).permute(1,0,2).reshape(5+80, -1)  # <--- how does this reshape work?
            scores += [item[4, :].sigmoid()]  # <-- why 4?

        return scores

Also referring to pytorch-YOLOv4, if you were to use their model how would you calculate the cls_scores? Their output is different from yours despite looking like the same functions.