Architecture and performance of pinholenet architectures.
<p><b>(A)</b> Schematic representation of PinholeNet architectures with varying receptive field sizes (P11, P17, P31, P33), highlighting the constrained receptive fields in the convolutional layers. For comparison, a standard AlexNet model is also shown in bottom. <b>(B)</...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | , |
| منشور في: |
2025
|
| الموضوعات: | |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | <p><b>(A)</b> Schematic representation of PinholeNet architectures with varying receptive field sizes (P11, P17, P31, P33), highlighting the constrained receptive fields in the convolutional layers. For comparison, a standard AlexNet model is also shown in bottom. <b>(B)</b> Top-1 object recognition accuracy on the ImageNet validation set during training for different PinholeNet models (P11, P17, P31, P33) and the standard AlexNet model. <b>(C)</b> Contour readout accuracy from the Conv5 layer (C5) for models fine-tuned on contour detection. The plot shows the performance of different PinholeNet models compared to the standard AlexNet model across finetuning epochs. <b>(D)</b> Receptive field sizes of intermediate units across different layers (C1 to the 2nd FC layer) in PinholeNet and standard AlexNet models. <b>(E)</b> Contour readout accuracy from other layers – the avgpool layer, 1st fully connected layer, and 2nd fully connected layer, for different PinholeNet models (P11, P17, P31, P33) and the standard AlexNet model. The legend denotes the color and line style for each model.</p> <p>(TIF)</p> |
|---|