restore_prefix error when trying to implement SWA

[kaiwangx]

Hi, all
I'm having an error when trying to implement the Stochastic Weight Averaging method
For simplicity, I replace the actual parameter with the pretrained yolo params.
I'm trying to calculate the average weight in each layer and save them as swa.params. I looked into the link in the error message but the solution there is replacing collect_params() into load and save params, which is not a solution for me because, in that way, I can not calculate the average.

Any idea will be appreciated! Thanks in advance

my code:

net = model_zoo.get_model('yolo3_darknet53_voc', pretrained=True)
net.save_parameters("1.params", deduplicate=True)
net.save_parameters("2.params", deduplicate=True)
param_names_list= ["1.params", "2.params"]
params = []
for param_name in param_names_list:
    net.load_parameters(param_name)
    params.append(net.collect_params())

for layer_name in params[0]:
    sum = 0
    for i in range(len(params)):
        current_param = params[i]
        sum += current_param[layer_name].data()
    params[0][layer_name].set_data(sum / len(params))
params[0].save("swa.params")

net = model_zoo.get_model('yolo3_darknet53_voc', pretrained=False)
net.load_parameters('swa.params')

Error message

restore_prefix is 'yolov32_' but Parameters name 'darknetv32_conv0_weight' does not start with 'yolov32_'. For more info on naming, please see https://mxnet.io/api/python/docs/tutorials/packages/gluon/blocks/naming.html