Record the performance metric during training

Keep track of most recent, average, sum, and count of a metric.

generation channel attention mask

generation spatial attention mask

A dict format repository to register module. Repository can also manage config node.

implementation of AP-CNN on ResNet

Channel Interaction Network Classifier

Channel Interaction Network

Compute compact bilinear pooling over two bottom inputs. Args: output_dim: output dimension for compact bilinear pooling.

Class for extracting activations and registering gradients from targetted intermediate layers

Convolutional layer that computes the squared L2 distance instead of the conventional inner product.

Matrix power normalized Covariance pooling (MPNCOV) implementation of fast MPN-COV (i.e.,iSQRT-COV) https://arxiv.org/abs/1712.01034

Class for making a forward pass, and getting: 1. The network output. 2. Activations from intermeddiate targetted layers. 3. Gradients fro

Multi smooth loss.

N-pairs loss as explained in equation 11 of MAMC paper. Reference: Multi-Attention Multi-Class Constraint for Fine-grained Image Recognit

Attention pyramid module with bottom-up attention pathway

Feature pyramid module with top-down feature pathway

Randomly apply Cutmix to the provided batch and targets. The class implements the data augmentations as described in the paper `"CutMix: Regul

Randomly apply Mixup to the provided batch and targets. The class implements the data augmentations as described in the paper `"mixup: Beyond

Test a model from a config which could be a training config.

Base trainer