output with shape [1, 448, 448] doesn't match the broadcast shape [3, 448, 448]

跑RACNN代码的时候有一个报错，改了几天还是没解决

RuntimeError: output with shape [1, 448, 448] doesn't match the broadcast shape [3, 448, 448]
问题相关代码
apn_iter, apn_epoch, apn_steps = pretrainAPN(trainset, trainloader)#运行到这一句出错

img

        images, labels = next(batch_iterator)

img

运行结果及报错内容

img

RuntimeError: output with shape [1, 448, 448] doesn't match the broadcast shape [3, 448, 448]

我应该是在trainer.py改吗？还是在它调用的函数改