在cuhk_sysu做完训练后使用demo.py推理结果不正确

[axia75]

感谢您提供重要的学习内容，有一个问题想请教，在cuhk_sysu做完训练后通过demo.py做推理的时候，结果不正确

训练了20轮
整个函数是不是需要做修改

我改为如下
def resume_from_ckpt11(ckpt_path, model, optimizer=None, lr_scheduler=None):
ckpt = torch.load(ckpt_path)
if 'model' in ckpt.keys():
ckpt = ckpt['model']

count = 0
miss = []
for i in ckpt:
    if 'backbone.swin' in i:          
        model.state_dict()[i].copy_(ckpt[i])          
        count += 1
    elif 'roi_heads.reid_head.swin' in i:
        model.state_dict()[i].copy_(ckpt[i])
        count += 1    
    else:
        miss.append(i)
print('%d loaded, %d missed:' %(count,len(miss)),miss)
return 0

不知道这么改对不对 或者应该怎么改

[axia75]

resume_from_ckpt 是不是需要修改

[AlecDusheck]

#3 for weights. you train on top of the SOLDIER pre-trained model.