[4_efficientdet] Raise KeyError during training in CPU

[SadiaAfrinPurba]

If I set use_gpu=false or CUDA is not available, then I got the following KeyError-

Traceback (most recent call last): File "D:\efficientdet\trainer.py", line 28, in train gtf.Train_Dataset(root_dir=self.dataset_root_dir, File "D:\train_detector.py", line 61, in Train_Dataset "batch_size": self.system_dict["params"]["batch_size"] * self.system_dict["local"]["num_gpus"], KeyError: 'num_gpus'

I tried to fix the problem by initializing these parameters in constructor- self.system_dict["local"]["num_gpus"], self.system_dict["local"]["device"]

class Detector():
    def __init__(self, verbose=1):
        self.system_dict = {}
        self.system_dict["verbose"] = verbose
        self.system_dict["local"] = {}
        self.system_dict["local"]["num_gpus"] = 1 <- set value as 1
        self.system_dict["local"]["device"] = 'cuda' if torch.cuda.is_available() else 'cpu' <- to avoid KeyError: 'device'`

train_detector.py

After that, I got the following errors-

Traceback (most recent call last):
  File "c:\users\appdata\local\programs\python\python38\lib\runpy.py", line 192, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "c:\users\appdata\local\programs\python\python38\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "D:\train_detector.py", line 197, in Train
    cls_loss, reg_loss = self.system_dict["local"]["model"](
  File "D:\site-packages\torch\nn\modules\module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "model.py", line 258, in forward
    c3, c4, c5 = self.backbone_net(img_batch)
  File "D:\site-packages\torch\nn\modules\module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "model.py", line 187, in forward
    x = self.model._swish(self.model._bn0(self.model._conv_stem(x)))
  File "site-packages\torch\nn\modules\module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "site-packages\efficientnet_pytorch\utils.py", line 144, in forward
    x = F.conv2d(x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

Is there any solution to train efficiendet-b0 model in CPU?

[abhi-kumar]

Yes, I think updates to a few places might make it work. I will check it up in a couple of days and update you on the same.