Solved EEP 596 Computer Vision HW #6

Description

1. (15 points) Number of parameters.  Write a function to compute the number of trainable parameters in a model, the function takes as input the model, e.g., ResNet-34.  Return the estimated number of parameters. The question will be automatically graded.

2. Global average pooling (30 points). Please briefly revisit HW4 before starting. Reuse the CIFAR10_dataset_a() implementation from the previous assignment for the dataset. Create a network with the following parameters:

GAPNet(

(conv1): Conv2d(3, 6, kernel_size=(5, 5), stride=(1, 1))

(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)

(conv2): Conv2d(6, 10, kernel_size=(5, 5), stride=(1, 1))

(gap): AvgPool2d(kernel_size=10, stride=10, padding=0)

(fc): Linear(in_features=10, out_features=10, bias=True)

)

Train the network for 10 epochs with learning rate =0.001 and momentum=0.9. Use Stochastic Gradient Descent as the optimizer.  Evaluate the model and save the weights as Gap_net_10epoch.pth and submit it on Gradescope.

1. Backbones (10 points). Download resnet18 (pretrained on ImageNet). Remove the final fully connected layer (classifier).  Return the features for “cat_eye.jpg”.
2. Transfer learning (30 points).  Download resnet18 (pretrained on ImageNet), modify the last layer to output 10 classes.  Freeze all the weights except the last layer.  (`for param in model.parameters(): param.requires_grad = False` and so forth.  See herㄨㄜe.)

Train the network for 10 epochs with learning rate =0.001 and momentum=0.9. Use Stochastic Gradient Descent as the optimizer.  Evaluate the model and save the weights as Res_net_10epoch.pth and submit it on Gradescope.

1. (25 points) DenseNet.  Write PyTorch code to implement MobileNet.  Derive your class from nn.Module, and implement both __init__() and forward() methods.  Be sure to include batch norm and ReLU.  Run your network on an image to be sure that all the dimensions are correct; you do not have to check that the output makes sense.  (It will not, since the network is not trained.) The question will be automatically graded.