Tata Consultancy Services Limited

Vivek Bangalore Sampathkumar

Balamuralidhar Purushothaman

Jayavardhana Rama Gubbi Lakshminarasimha

Vartika Sengar

Arpan Pal

Avik Ghose

Murali Poduval

Karthik Seemakurthy

Aparna Kanakatte Gurumurthy

Srinivasan Jayaraman

Rajan Mindigal Alasingara Bhattachar

Bagya Lakshmi Vasudevan

Gaurab Bhattacharya

Jayeeta Saha

Srinivsan Jayaraman

The disclosure generally relates to scene graph generation. Scene graph captures rich semantic information of an image by representing objects and their relationships as nodes and edges of a graph and has several applications including image retrieval, action recognition, visual question answering, autonomous driving, robotics. However, to leverage scene graphs, computationally efficient scene graph generation methods are required, which is very challenging to generate due presence of a quadratic number of potential edges and computationally intensive/non-scalable techniques for detecting the relationship between each object pair using the traditional approach. The disclosure proposes a combination of edge proposal neural network and the Graph neural network with spatial message passing (GNN-SMP) along with several techniques including a feature extraction technique, object detection technique, un-labelled graph generation technique and a scene graph generation technique to generate scene graphs.

Method and system for scene graph generation

This disclosure relates generally to identification and mitigation of bias while training deep learning models. Conventional methods do not provide effective methods for bias identification, and they require pre-defined concepts and rules for bias mitigation. The embodiments of the present disclosure train an auto-encoder to produce a generalized representation of an input image by decomposing into a set of latent embedding. The set of latent embedding are used to learn the shape and color concepts of the input image. The feature specialization is done by training an auto-encoder to reconstruct the input image using the shape embedding modulated by color embedding. To identify the bias, permutation invariant neural network is trained for classification task and attribution scores corresponding to each concept embedding are computed. The method also performs de-biasing the classifier by training it with a set of counterfactual images generated by modifying the latent embedding learned by the auto-encoder.

Method and system for identifying and mitigating bias while training deep learning models

The disclosure herein relates to methods and systems for generating an end-to-end de-smoking model for removing smoke present in a video. Conventional data-driven based de-smoking approaches are limited mainly due to lack of suitable training data. Further, the conventional data-driven based de-smoking approaches are not end-to-end for removing the smoke present in the video. The de-smoking model of the present disclosure is trained end-to-end with the use of synthesized smoky video frames that are obtained by source aware smoke synthesis approach. The end-to-end de-smoking model localize and remove the smoke present in the video, using dynamic properties of the smoke. Hence the end-to-end de-smoking model simultaneously identifies the regions affected with the smoke and performs the de-smoking with minimal artifacts. localized smoke removal and color restoration of a real-time video.

Methods and systems for generating end-to-end de-smoking model

The disclosure herein relates to methods and systems for localized smoke removal and color restoration of a real-time video. Conventional techniques apply the de-smoking process only on a single image, by finding the regions having the smoke, based on manual air-light estimation. In addition, regaining original colors of de-smoked image is quite challenging. The present disclosure herein solves the technical problems. In the first stage, video frames having the smoky and smoke-free video frames are identified, from the video received in the real-time. In the second stage, an air-light is estimated automatically using a combined feature map. An intermediate de-smoked video frame for each smoky video frame is generated based on the air-light using a de-smoking algorithm. In the third and the last stage, a smoke-free video reference frame is used to compensate for color distortions introduced by the de-smoking algorithm in the second stage.

Methods and systems for localized smoke removal and color restoration of real-time video

Growing community of inventors

Bengaluru, India

Vivek Bangalore Sampathkumar