FlexConv: Continuous Kernel Convolutions with Differentiable Kernel Sizes

Benchmark Model Rank Results
sequential-image-classification-on-noise-padded-cifar-10FlexTCN-6#1% Test Accuracy: 69.87%
sequential-image-classification-on-sequential-mnistFlexTCN-4#3Permuted Accuracy: 98.72%
sequential-image-classification-on-sequential-mnistFlexTCN-6#28Unpermuted Accuracy: 99.62%
time-series-on-speech-commandsFlexTCN-4#3% Test Accuracy: 97.73
time-series-on-speech-commandsFlexTCN-6#6% Test Accuracy (Raw Data): 91.73
sequential-image-classification-on-sequential-cifar-10FlexTCN-6#6Unpermuted Accuracy: 80.82%
image-classification-on-cifar-10FlexTCN-7#216Percentage correct: 92.2