WindVChen

哈喽 @Snowbluerose ， 如果指的是给输入图片打标签的话，可以利用`--save-txt`参数把预测的标签存下来。

我们是自己写了一个python打标程序，你也可以采用[labelme](https://github.com/wkentaro/labelme)来快速打标

标签格式对标的是YOLOv5的标签格式 (class_id center_x center_y width height)，可以参考下图，以及YOLOv5的[官方手册介绍](https://roboflow.com/formats/yolov5-pytorch-txt?ref=ultralytics) 

Hi @ramdhan1989 , Sorry for replying late. You can modify the codeline in Line121, yolo.py from this: ``` yi = self.forward_once(xi)[0] # forward ``` to this: ``` yi = self.forward_once(xi)[0][0]...

## PROBLEM Actually you only need to change the following part: ``` class BottleneckResAtnMHSA(nn.Module): # Standard bottleneck def __init__(self, n_dims, size, shortcut=True): # ch_in, ch_out, shortcut, groups, expansion super(BottleneckResAtnMHSA, self).__init__()...

You can consider interpolations on the output of the following codeline: ``` def forward(self, x): ... content_position = (self.rel_h + self.rel_w).view(1, C, -1).permute(0, 2, 1) ... ``` From the Table...

Modifications: ``` Class BottleneckResAtnMHSA(nn.Module): ... # content_position = (self.rel_h + self.rel_w).view(1, C, -1).permute(0, 2, 1) content_position = (self.rel_h + self.rel_w) content_position = nn.functional.interpolate(content_position, (int(content_content.shape[-1]**0.5), int(content_content.shape[-1]**0.5)), mode='bilinear') content_position = content_position.view(1, C,...

You're right. The saying "train the model using different size of image" may just mean that we don't need to recalculate the parameters in DRENet.yaml for input images with sizes...

Hi @tommylatranvu , This is a problem caused by an incompatible torch version. You can find some discussions about this [here](https://github.com/ultralytics/yolov5/issues/8405). And there are some simple and easy ways online...

Hi @namita-agarwal , 1. Is your data structure consistent with [here](https://github.com/WindVChen/DRENet/issues/2#issuecomment-1236038070)? 2. And the dataset path in the `ship.yaml` may need to target to **images** subfolder, not **degrade**. You can...