使用ssd_notebook.ipynb测试自己的模型出现问题,求大神解答?

[zhangfudiyi]

在加载自己训练的模型时,出现错误 InvalidArgumentError : Assign requires shapes of both tensors to match. lhs shape= [126] rhs shape= [12]. 希望能得到解答，不知道为啥。谢了~！

[whmin]

我也遇到了同样的问题.....

[zhangfudiyi]

问题已经解决，源代码中ssd_vgg_300.py中分类数需要修改和你自己数据一样的分类数目，作者给的是个指定值21。

[whmin]

我也发现了,谢谢! 你训练的loss正常收敛了吗?

[KKKlose]

您好，我加载自己的模型，报错： NotFoundError (see above for traceback): Restoring from checkpoint failed. This is most likely due to a Variable name or other graph key that is missing from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:

Key ssd_300_vgg/conv1/conv1_1/biases/Adam not found in checkpoint 这个怎么解决？

[dayL-W]

@KKKlose 你加载自己的模型需要保证网络结构和名称都是对的，不然就加载失败

[speculaas]

Dear All, Thanks for sharing this issue.

At first I follow the advice here, trying to modify ssd_vgg_300.py, but I got different error which was also lhs, rhs not matching.

Then, looking at how ssd_net is declared in train_ssd_network.py, I add the following code in jupyter notebook, and was able to run the notebook without error

ssd_class = ssd_vgg_300.SSDNet ssd_params = ssd_class.default_params._replace(num_classes=15) ssd_net = ssd_class(ssd_params)

BR, JimmyYS

[2017TJM]

要怎么解决了，InvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [12] rhs shape= [126]

[2017TJM]

已经解决了，但是模型测试精度不行

[imilesmile]

已经解决了，但是模型测试精度不行

麻烦问下怎么解决的呢?

[yuchanWang]

已经解决了，但是模型测试精度不行 我在做eval时，遇到了类似的问题，但我的ssd_vgg_300.py中的num_class已经修改过了还是出现这个问题。 请问能告知一下您的方法吗？我真的急切需要，拜托了!

[DalalAlmutawa]

change line 96 at nets/ssd_vgg_300.py to num_classes= your custom class number