Downloading Resnet50 onto IDLE takes Long TImes

TensorFlow
, , ,
1

Hi,

For a science fair project, I am using transfer learning for Resnet 50. I am having some issues loading it, so I tried downloading basic Resnet50 with random weights (though I plan on trying image weights later).

The code I’m running is:

base_cnn = ResNet50()

My output is as follows

8192/102967424 […] - ETA: 0s
16384/102967424 […] - ETA: 10:01
24576/102967424 […] - ETA: 11:22
32768/102967424 […] - ETA: 12:50
40960/102967424 […] - ETA: 13:47
49152/102967424 […] - ETA: 14:22
57344/102967424 […] - ETA: 14:48
65536/102967424 […] - ETA: 15:08
73728/102967424 […] - ETA: 15:24
81920/102967424 […] - ETA: 15:36
450560/102967424 […] - ETA: 3:05 
598016/102967424 […] - ETA: 2:27
712704/102967424 […] - ETA: 2:11
860160/102967424 […] - ETA: 1:54
868352/102967424 […] - ETA: 2:00
876544/102967424 […] - ETA: 2:07
884736/102967424 […] - ETA: 2:14
892928/102967424 […] - ETA: 2:22
901120/102967424 […] - ETA: 2:30
909312/102967424 […] - ETA: 2:40
917504/102967424 […] - ETA: 2:51
925696/102967424 […] - ETA: 3:05
933888/102967424 […] - ETA: 3:19
942080/102967424 […] - ETA: 3:36
950272/102967424 […] - ETA: 3:53
958464/102967424 […] - ETA: 4:13
966656/102967424 […] - ETA: 4:35
974848/102967424 […] - ETA: 4:59
983040/102967424 […] - ETA: 5:26
991232/102967424 […] - ETA: 5:56
999424/102967424 […] - ETA: 6:28
1007616/102967424 […] - ETA: 7:02
1015808/102967424 […] - ETA: 7:41

This is being run on IDLE 3.10.1, and the time stamps are increasing very rapidly and IDLE lags very quickly following this. The laptop is not the issue; I have a Macbook Pro from 2 years ago. I have udpated tensorflow as well. Some reason resnet installs very quickly on google colab but I can’t use this as google colab doesn’t seem to be able to handle my training data images very well (which in comparison works better on IDLE).

Any immediate help is greatly appreciated,

Best,

Arnav

3

Hi @adagar ,

In general IDLE itself is not optimized for heavy computational tasks.The increasing timestamps and rapid IDLE lagging indicate that downloading and potentially loading the ResNet50 model are resource-intensive and your laptop might not have sufficient CPU or RAM for complex deep learning tasks.

Could you please try the following points and let us know if it is improve the downloading process:

1.Use smaller architecture models Instead of ResNet50, consider using a smaller pre-trained model like MobileNetV2 or VGG16, which have fewer parameters and require less processing power to load.
2. If you need ResNet50's architecture, use a pre-trained model with ImageNet weights instead of random weights, as pre-trained weights potentially load faster.
3.If you have a GPU available, enable it in your TF configuration to offload computations from your CPU.

Hope these points will helpful.

Thanks