ValueError: not enough values to unpack (expected 2, got 1)


Hi everyone,
I am doing mxnet rcnn trainign on my own data.
my data is of 512x512x3.
i am using mobilenet instead of vgg or resnet as a backbone.
i have added a symbol file in rcnn with mobilenet netowork.
i have params file of mobilenet0.25…
but when i run the training script with network as mobilenet,
i am getting this below error.

Traceback (most recent call last):
File “”, line 195, in
File “”, line 192, in main, lr_step=args.lr_step)
File “”, line 84, in train_net
arg_params, aux_params = load_param(pretrained, epoch, convert=True)
File “/home/ubuntu/incubator-mxnet/example/rcnn/rcnn/utils/”, line 66, in load_param
arg_params, aux_params = load_checkpoint(prefix, epoch)
File “/home/ubuntu/incubator-mxnet/example/rcnn/rcnn/utils/”, line 36, in load_checkpoint
tp, name = k.split(’:’, 1)
ValueError: not enough values to unpack (expected 2, got 1)

what could be the problem here?
could you guys help me out in understanding this problem… thank you.


Hi @Ram124,

You seem to have a parameter file that’s not formatted in the way expected by the function. Could you share a link to the MobileNet-0.25 parameter file, so I can take a look. Or are you able to inspect the key to check that it is in “type:name” format, as expected by this function?


I have used the gluon mobilenet-0.25 param file.
I have not checked it…
I did not understand what it is exactly…
Gluon model does not support?


As with all models, MobileNet-0.25 contains a number of learnable parameters (the kernels, biases, weights, etc). Someone else has trained a MobileNet-0.25 model and saved the values of all their parameters after training, so you don’t have to train your own model. You then just create a MobileNet-0.25 network, and replace all of the parameters in your model with these pre-trained parameters: no training required!

BUT the parameters file must be in a specific format, and this is what I’d like to check. Where did you get this params file from? And Gluon does support this functionality.