sdtemple/shape-prediction-model

This model predicts the shape (circle, rectangle, diamond, or triangle) of the 1 colored shape (8 colors) in a 224 x 224 x 3 image.

This model is a part of a how to tutorial on fitting PyTorch models.

The model is trained on 2000 examples for each color and shape combo (64,000 samples in total) simulated according to https://github.com/sdtemple/zootopia3.

The model is tested/evaluated on the dataset https://huggingface.co/datasets/sdtemple/colored-shapes, which has slightly smaller shapes simulated (out of distribution) relative to the training data. The metrics below can be +- a few points depending on random seed.

• Accuracy: 75%
• Min precision (triangle): 57%
• Max precision (rectangle): 98%
• Min recall (diamond): 66%
• Max recall (triangle): 84%
• AUROC (macro-averaged): 92%
• Min AUROC (diamond): 90%
• Max AUROC (circle): 94%

Compared to https://huggingface.co/sdtemple/color-prediction-model, it is harder to predict the shape than the color of the object.

The model architecture is the following. In light experimentation, I found it important to have multiple convolutions and that too many parameters leads to noisy validation losses by epoch.

MyCNN(
  (conv_block): Sequential(
    (0): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (3): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (4): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (6): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (7): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (8): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (9): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (10): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (11): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (12): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (14): AvgPool2d(kernel_size=2, stride=2, padding=0)
  )
  (linear_block): Sequential(
    (0): Linear(in_features=784, out_features=16, bias=True)
    (1): BatchNorm1d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
    (3): Dropout(p=0.2, inplace=False)
    (4): Linear(in_features=16, out_features=16, bias=True)
    (5): BatchNorm1d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (6): ReLU()
    (7): Dropout(p=0.2, inplace=False)
  )
  (output_block): Linear(in_features=16, out_features=4, bias=True)
)