Adobe, Inc.

Jason Wen Yong Kuen

Zhe Lin

Jianming Zhang

Scott David Cohen

Xiaohui Shen

Mingyang Ling

Jiuxiang Gu

Handong Zhao

Vlad Ion Morariu

John Philip Collomosse

Zhifei Zhang

Tong Sun

Sijie Zhu

Xin Yuan

Brian Lynn Price

Ajinkya Gorakhnath Kale

Baldo Antonio Faieta

Yilin Wang

Rajiv Jain

Simon Chen

Qing Liu

Zijun Wei

Nikolaos Barmpalios

Su Chen

Ani Nenkova

Dat Ba Huynh

Bo Sun

Yu Zeng

Sukjun Hwang

Embodiments of the present disclosure include obtaining a text prompt describing an element, layout information indicating a target region for the element, and a precision level corresponding to the element. Some embodiments generate a text feature pyramid based on the text prompt, the layout information, and the precision level, wherein the text feature pyramid comprises a plurality of text feature maps at a plurality of scales, respectively. Then, an image is generated based on the text feature pyramid. In some cases, the image includes an object corresponding to the element of the text prompt at the target region. Additionally, a shape of the object corresponds to a shape of the target region based on the precision level.

Semantic image synthesis

Systems and methods for image processing are described. Embodiments of the present disclosure receive a training image and a caption for the training image, wherein the caption includes text describing an object in the training image; generate a pseudo mask for the object using a teacher network based on the text describing the object; generate a mask for the object using a student network; and update parameters of the student network based on the mask and the pseudo mask.

Open vocabulary instance segmentation

Methods and systems disclosed herein relate generally to systems and methods for generating visual relationship graphs that identify relationships between objects depicted in an image. A vision-language application uses transformer encoders to generate a graph structure, in which the graph structure represents a dependency between a first region and a second region of an image. The dependency indicates that a contextual representation of the first region was derived, at least in part, by processing the second region. The contextual representation identifies a predicted identity of an image object depicted in the first region. The predicted identity is determined at least in part by identifying a relationship between the first region and other data objects associated with various modalities.

Self-supervised visual-relationship probing

In implementations of systems for performing multiple segmentation tasks, a computing device implements a segment system to receive input data describing a digital image depicting an object. The segment system computes per-pixel embeddings for the digital image using a pixel decoder of a machine learning model. Output embeddings are generated using a transformer decoder of the machine learning model based on the per-pixel embeddings for the digital image, input embeddings for a first segmentation task and input embeddings for a second segmentation task. The segment system outputs a first digital image and a second digital image. The first digital image depicts the object segmented based on the first segmentation task and the second digital image depicts the object segmented based on the second segmentation task.

Performing multiple segmentation tasks

The present disclosure relates to systems, methods, and non-transitory computer readable media that utilizes artificial intelligence to learn to recommend foreground object images for use in generating composite images based on geometry and/or lighting features. For instance, in one or more embodiments, the disclosed systems transform a foreground object image corresponding to a background image using at least one of a geometry transformation or a lighting transformation. The disclosed systems further generating predicted embeddings for the background image, the foreground object image, and the transformed foreground object image within a geometry-lighting-sensitive embedding space utilizing a geometry-lighting-aware neural network. Using a loss determined from the predicted embeddings, the disclosed systems update parameters of the geometry-lighting-aware neural network. The disclosed systems further provide a variety of efficient user interfaces for generating composite digital images.

Recommending objects for image composition using geometry-and-lighting aware search and efficient user interface workflows

Recommending objects for image composition using a geometry-and-lighting aware neural network

The technology described includes methods for pretraining a document encoder model based on multimodal self cross-attention. One method includes receiving image data that encodes a set of pretraining documents. A set of sentences is extracted from the image data. A bounding box for each sentence is generated. For each sentence, a set of predicted features is generated by using an encoder machine-learning model. The encoder model performs cross-attention between a set of masked-textual features for the sentence and a set of masked-visual features for the sentence. The set of masked-textual features is based on a masking function and the sentence. The set of masked-visual features is based on the masking function and the corresponding bounding box. A document-encoder model is pretrained based on the set of predicted features for each sentence and pretraining tasks. The pretraining tasks includes masked sentence modeling, visual contrastive learning, or visual-language alignment.

Unified pretraining framework for document understanding

The present disclosure relates to systems, methods, and non-transitory computer-readable media that generates object masks for digital objects portrayed in digital images utilizing a detection-masking neural network pipeline. In particular, in one or more embodiments, the disclosed systems utilize detection heads of a neural network to detect digital objects portrayed within a digital image. In some cases, each detection head is associated with one or more digital object classes that are not associated with the other detection heads. Further, in some cases, the detection heads implement multi-scale synchronized batch normalization to normalize feature maps across various feature levels. The disclosed systems further utilize a masking head of the neural network to generate one or more object masks for the detected digital objects. In some cases, the disclosed systems utilize post-processing techniques to filter out low-quality masks.

Detecting digital objects and generating object masks on device

Systems and methods for multi-modal representation learning are described. One or more embodiments provide a visual representation learning system trained using machine learning techniques. For example, some embodiments of the visual representation learning system are trained using cross-modal training tasks including a combination of intra-modal and inter-modal similarity preservation objectives. In some examples, the training tasks are based on contrastive learning techniques.

Visual-semantic representation learning via multi-modal contrastive training

Systems and methods for image generation are described. Embodiments of the present disclosure receive a text phrase that describes a target image to be generated; generate text features based on the text phrase; retrieve a search image based on the text phrase; and generate the target image using an image generation network based on the text features and the search image.

Retrieval-based text-to-image generation with visual-semantic contrastive representation

Systems and methods for image processing are described. Embodiments of the present disclosure receive an image having a plurality of object instances; encode the image to obtain image features; decode the image features to obtain object features; generate object detection information based on the object features using an object detection branch, wherein the object detection branch is trained based on a first training set using a detection loss; generate semantic segmentation information based on the object features using a semantic segmentation branch, wherein the semantic segmentation branch is trained based on a second training set different from the first training set using a semantic segmentation loss; and combine the object detection information and the semantic segmentation information to obtain panoptic segmentation information that indicates which pixels of the image correspond to each of the plurality of object instances.

Multi-source panoptic feature pyramid network

In implementations of object detection in images, object detectors are trained using heterogeneous training datasets. A first training dataset is used to train an image tagging network to determine an attention map of an input image for a target concept. A second training dataset is used to train a conditional detection network that accepts as conditional inputs the attention map and a word embedding of the target concept. Despite the conditional detection network being trained with a training dataset having a small number of seen classes (e.g., classes in a training dataset), it generalizes to novel, unseen classes by concept conditioning, since the target concept propagates through the conditional detection network via the conditional inputs, thus influencing classification and region proposal. Hence, classes of objects that can be detected are expanded, without the need to scale training databases to include additional classes.

Object detection in images

The present disclosure relates to systems, methods, and non-transitory computer readable media for accurately and efficiently learning parameters of a distilled neural network from parameters of a source neural network utilizing multiple augmentation strategies. For example, the disclosed systems can generate lightly augmented digital images and heavily augmented digital images. The disclosed systems can further learn parameters for a source neural network from the lightly augmented digital images. Moreover, the disclosed systems can learn parameters for a distilled neural network from the parameters learned for the source neural network. For example, the disclosed systems can compare classifications of heavily augmented digital images generated by the source neural network and the distilled neural network to transfer learned parameters from the source neural network to the distilled neural network via a knowledge distillation loss function.

Growing community of inventors

San Jose, CA, United States of America

Jason Wen Yong Kuen