little container class for holding the detection result od: object detector, rm: rel model

Visualizer that draws data about detection/segmentation on images. It contains methods like `draw_{text,box,circle,line,binary_mask,polygon}`

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

from root to leaves.

From leaves to root.

Bidirectional Tree LSTM Contains one forward lstm(leaves to root) and one backward lstm(root to leaves) Dropout mask will be generated one time

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

Very simple multi-layer perceptron (also called FFN) Copied from hoitr.

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

Multilayer Bidirectional Tree LSTM Each layer contains one forward lstm(leaves to root) and one backward lstm(root to leaves)

One Way Tree LSTM direction = forward | backward

Adapted from the [CVPR 2020] GPS-Net: Graph Property Scensing Network for Scene Graph Generation]

The goal of this is to provide a simplified way of computing P(predicate. | obj1, obj2, img).

Modified from neural-motifs to encode contexts for each objects.

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

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

Very simple multi-layer perceptron (also called FFN) Copied from hoitr.

Very simple multi-layer perceptron (also called FFN) Copied from hoitr.

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

Obtain the final relation information for evaluation.

Modified from neural-motifs to encode contexts for each objects.

This API is used for obtaining the final information for demonstrating the scene graphs. It's usually invoked after the PostProcessor. Especi

Modify the DETR transformer with two decoders. Args: encoder (`mmcv.ConfigDict` | Dict): Config of TransformerEncoder. Defaul

Random crop the image & bboxes & masks & scene relations.

The basic class of all the relation head.

Transfer gt_rels to tensor too.

Extract RoI features from a single level feature map. If there are multiple input feature levels, each RoI is mapped to a level according to