This module converts the model's output into the format expected by the coco api

Very simple multi-layer perceptron (also called FFN)

Track a series of values and provide access to smoothed values over a window or the global series average.

r"""Allows the model to jointly attend to information from different representation subspaces. See reference: Attention Is All You Need .

Similar to PostProcess but for segmentation masks. This processor is to be called sequentially after PostProcess. Args: threshold: t

ResNet backbone with frozen BatchNorm.

This is the MDETR module that performs modulated object detection

This is a more standard version of the position embedding, very similar to the one used by the Attention is all you need paper, generalized t

Optimized union find structure

Dataset for eval only. Provides the question and the image

This class takes as input a set of embeddings of dimension C1 and outputs a set of embedding of dimension C2, after a linear transformation,

BatchNorm2d where the batch statistics and the affine parameters are fixed. Copy-paste from torchvision.misc.ops with added eps before rqsrt

ResNet backbone with GroupNorm with 32 channels.

This class computes an assignment between the targets and the predictions of the network For efficiency reasons, the targets don't include the no

This is a 2D attention module, which only returns the attention softmax (no multiplication by value)

Very simple multi-layer perceptron (also called FFN)

Same as the modulated detection dataset, except with multiple img sources

Absolute pos embedding, learned.

This module converts the model's output for Flickr30k entities evaluation. This processor is intended for recall@k evaluation with respect to eac

This module converts the model's output for Phrasecut evaluation. Contrary to most postprocessors, it doesn't operate directly on model's output,

Utility class to track recall@k for various k, split by categories

This class computes the loss for DETR. The process happens in two steps: 1) we compute hungarian assignment between ground truth boxes and

Optimized union find structure

Coco-style dataset imported from TorchVision. It is modified to handle several image sources Args: root_coco (string): Path to the c

Simple convolutional head, using group norm. Upsampling is done using a FPN approach