loading pretrained model

[jawaechan]

Hi,
I download your trained model BiDet-SSD300-VOC_66.0.pth and want to load it with the network in bidet.ssd.py. But I met an issue of mismatching size as follows:

RuntimeError: Error(s) in loading state_dict for BiDetSSD:
size mismatch for conf.0.weight: copying a param with shape torch.Size([84, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([324, 512, 3, 3]).
size mismatch for conf.0.bias: copying a param with shape torch.Size([84]) from checkpoint, the shape in current model is torch.Size([324]).
size mismatch for conf.1.weight: copying a param with shape torch.Size([126, 1024, 3, 3]) from checkpoint, the shape in current model is torch.Size([486, 1024, 3, 3]).
size mismatch for conf.1.bias: copying a param with shape torch.Size([126]) from checkpoint, the shape in current model is torch.Size([486]).
size mismatch for conf.2.weight: copying a param with shape torch.Size([126, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([486, 512, 3, 3]).
size mismatch for conf.2.bias: copying a param with shape torch.Size([126]) from checkpoint, the shape in current model is torch.Size([486]).
size mismatch for conf.3.weight: copying a param with shape torch.Size([126, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([486, 256, 3, 3]).
size mismatch for conf.3.bias: copying a param with shape torch.Size([126]) from checkpoint, the shape in current model is torch.Size([486]).
size mismatch for conf.4.weight: copying a param with shape torch.Size([84, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([324, 256, 3, 3]).
size mismatch for conf.4.bias: copying a param with shape torch.Size([84]) from checkpoint, the shape in current model is torch.Size([324]).
size mismatch for conf.5.weight: copying a param with shape torch.Size([84, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([324, 256, 3, 3]).
size mismatch for conf.5.bias: copying a param with shape torch.Size([84]) from checkpoint, the shape in current model is torch.Size([324]).
Can you advise how to fix it?

Thanks

[samagalhaes]

Hi,
I am using it now and looks fine. How are you loading the weights? Are you using the internal function?

ssd = build_bidet_ssd('test', 300, 21, nms_conf_thre=0.03)

model_path=build_dir + "/pretrain/BiDet-SSD300-VOC_66.0.pth"
ssd.load_weights(model_path)

[jawaechan]

Hi,
I use the following codes:

model = build_bidet_ssd('train', 300, 80,
nms_conf_thre=0.03)
vgg_weights = torch.load('./pretrained/BiDet-SSD300-VOC_66.0.pth')
model.load_state_dict(vgg_weights,strict=True)

Thanks

[samagalhaes]

Hi,

That should also work. I think you need to set strict to False. Anyway the best is to use the load function.

[jawaechan]

I tried your code. I find when the parameter num_class is set as 21, there is no problem. But I met the error when it is set as 80.

Thanks

[samagalhaes]

Mind that the pre-trained SSD was trained in the PASCAL VOC dataset, which only has 21 classes. To use the COCO dataset, you need to train your own network.

[Wuziyi616]

@samagalhaes Thanks for your kind reply, and yes @Jawae the pre-trained model is for VOC which has 21 classes, it can't be used for COCO which has 81 classes.

[jawaechan]

The final question: can I know what is the loss value when the training converges?

[Wuziyi616]

Sorry for my late reply. If I remember correctly, on VOC dataset, the loc_loss+cls_loss is around 5, with about 1.6 and 3.2 for each (sorry I can't remember exactly which belongs to which loss term)

[Wuziyi616]

But actually loss value doesn't mean final mAP, I will refer you to this issue.