公开(公告)号：US12198453B2

公开(公告)日：2025-01-14

申请号：US17587161

申请日：2022-01-28

Applicant: Salesforce, Inc.

Inventor： Mingfei Gao ,  Chen Xing

IPC: G06V20/62 ,  G06F40/126 ,  G06T1/60 ,  G06T9/00 ,  G06V10/22 ,  G06V10/77 ,  G06V10/774

Abstract: Embodiments described herein provide methods and systems for open vocabulary object detection of images. given a pre-trained vision-language model and an image-caption pair, an activation map may be computed in the image that corresponds to an object of interest mentioned in the caption. The activation map is then converted into a pseudo bounding-box label for the corresponding object category. The open vocabulary detector is then directly supervised by these pseudo box-labels, which enables training object detectors with no human-provided bounding-box annotations.

公开(公告)号：US20240070394A1

公开(公告)日：2024-02-29

申请号：US18160967

申请日：2023-01-27

Applicant: Salesforce, Inc.

Inventor： Xiangyu Peng ,  Chen Xing ,  Prafulla Kumar Choubey ,  Chieng-Sheng Wu

IPC: G06F40/284 ,  G06F40/40

CPC classification number: G06F40/284 ,  G06F40/40

Abstract: Embodiments described herein provide a mechanism that ensembles trainable soft prompts to transfer knowledge from source tasks under few-shot learning settings. Specifically, given a source task input from a source task training dataset, a set of soft prompts may be trained using a frozen PLM on the large-scale source task training dataset. The set of soft prompts are then prepended to a target task input, based on which the frozen pre-trained language model generates a set of logits for predicting classification of the target task input, respectively. An attention module is used to generate input-logit attention scores, which are used to compute a weighted linear combination of the logits given the attention scores. The weighted linear combination are the final logits to predict the final classification of the target task input.

公开(公告)号：US20240185035A1

公开(公告)日：2024-06-06

申请号：US18162535

申请日：2023-01-31

Applicant: Salesforce, Inc.

Inventor： Ning Yu ,  Can Qin ,  Chen Xing ,  Shu Zhang ,  Stefano Ermon ,  Caiming Xiong ,  Ran Xu

IPC: G06N3/0455 ,  G06T5/00

CPC classification number: G06N3/0455 ,  G06T5/002 ,  G06T2207/20084

Abstract: Embodiments described herein provide a mechanism for replacing existing text encoders in text-to-image generation models with more powerful pre-trained language models. Specifically, a translation network is trained to map features from the pre-trained language model output into the space of the target text encoder. The training preserves the rich structure of the pre-trained language model while allowing it to operate within the text-to-image generation model. The resulting modularized text-to-image model receives prompt and generates an image representing the features contained in the prompt.

公开(公告)号：US20240070868A1

公开(公告)日：2024-02-29

申请号：US18159318

申请日：2023-01-25

Applicant: Salesforce, Inc.

Inventor： Ning Yu ,  Vibashan Vishnukumar Sharmini ,  Chen Xing ,  Juan Carlos Niebles Duque ,  Ran Xu

IPC: G06T7/11 ,  G06V10/26

CPC classification number: G06T7/11 ,  G06V10/273

Abstract: Embodiments described herein provide an open-vocabulary instance segmentation framework that adopts a pre-trained vision-language model to develop a pipeline in detecting novel categories of instances.

公开(公告)号：US20230092702A1

公开(公告)日：2023-03-23

申请号：US17933396

申请日：2022-09-19

Applicant: Salesforce, Inc.

Inventor： Yixin Mao ,  Tian Xie ,  Chaney Lin ,  Chen Xing ,  Zachary Alexander ,  Wenhao Liu

IPC: G06F40/35 ,  G06F16/35 ,  G06F16/383

Abstract: Database systems and methods are provided for assigning structural metadata to records and creating automations using the structural metadata. One method of assigning structural metadata to a group of records involves determining, based on one or more fields of metadata associated with the records, a plurality of candidate names, wherein each candidate name of the plurality of candidate names corresponds to semantic content of the one or more fields of a respective record of the group of records, for each candidate name, assigning a name relevance score based on respective word relevance scores assigned to respective words of the respective candidate name based on usage, selecting a candidate name in a manner that is influenced by the respective name relevance scores assigned to the respective candidate names and automatically assigning a name to the group of records using the candidate name.

公开(公告)号：US20240330409A1

公开(公告)日：2024-10-03

申请号：US18738628

申请日：2024-06-10

Applicant: Salesforce, Inc.

Inventor： Chen Xing ,  Wenhao Liu ,  Chu Hong Hoi ,  Nitish Shirish Keskar ,  Caiming Xiong

IPC: G06F18/214 ,  G06F18/21 ,  G06F40/00

CPC classification number: G06F18/2148 ,  G06F18/2163 ,  G06F40/00

Abstract: Embodiments are directed to pre-training a transformer model using more parameters for sophisticated patterns (PSP++). The transformer model is divided into a held-out model and a main model. A forward pass and a backward pass are performed on the held-out model, where the forward pass determines self-attention hidden states of the held-out model and the backward pass determines loss of the held-out model. A forward pass on the main model is performed to determine a self-attention hidden states of the main model. The self-attention hidden states of the main model are concatenated with the self-attention hidden states of the held-out model. A backward pass is performed on the main model to determine a loss of the main model. The parameters of the held-out model are updated to reflect the loss of the held-out model and parameters of the main model are updated to reflect the loss of the main model.