SEARCH SESSIONS

Search All
 
Refine Results:
 
Year(s)

SOCIAL MEDIA

EMAIL SUBSCRIPTION

 
 

GTC On-Demand

Presentation
Media
Abstract:
We'll discuss learning to synthesize object instances such as a person or car in both 2D and 3D scenes. We will introduce our work we presented at NeurIPS 2018 on context-aware synthesis and placement of object instances. We propose a generative model that learns to generate and insert an object instance into an image in a semantically coherent manner. In particular, we represent object instances using masks and learn to insert them into semantic label maps of images. Our talk will also cover our recent work around putting humans in a scene and learning affordance in 3D indoor environments. This extends the learning of context from 2D to 3D scenes in which the synthesized objects are semantically coherent and geometrically correct. We'll show that both projects add technical insights and have potential applications in content creation.
We'll discuss learning to synthesize object instances such as a person or car in both 2D and 3D scenes. We will introduce our work we presented at NeurIPS 2018 on context-aware synthesis and placement of object instances. We propose a generative model that learns to generate and insert an object instance into an image in a semantically coherent manner. In particular, we represent object instances using masks and learn to insert them into semantic label maps of images. Our talk will also cover our recent work around putting humans in a scene and learning affordance in 3D indoor environments. This extends the learning of context from 2D to 3D scenes in which the synthesized objects are semantically coherent and geometrically correct. We'll show that both projects add technical insights and have potential applications in content creation.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2019
Session ID:
S9959
Streaming:
Download:
Share:
 
Abstract:
We provide a unified framework on learning affinity in pure data-driven fashion using a linear propagation structure. This is a GPU and deep learning friendly pairwise learning module that does not require solving linear equation, iterative inferences or manually defined kernels. Specifically, we develop a three-way connection for the linear propagation model, which formulates a sparse transformation matrix, where all elements can be the output from a deep CNN, but results in a dense affinity matrix that effectively models any task-specific pairwise similarity matrix. The spatial propagation network can be applied to many affinity-related tasks, such as image matting, segmentation and colorization, to name a few. Essentially, the model can learn semantically aware affinity relations for high-level vision tasks due to the powerful learning capability of the deep CNN. We validate the framework on the task of refinement for image segmentation boundaries. Experiments on face parsing and semantic segmentation tasks show that the spatial propagation network provides a general, effective, and efficient solution for generating high-quality segmentation results.
We provide a unified framework on learning affinity in pure data-driven fashion using a linear propagation structure. This is a GPU and deep learning friendly pairwise learning module that does not require solving linear equation, iterative inferences or manually defined kernels. Specifically, we develop a three-way connection for the linear propagation model, which formulates a sparse transformation matrix, where all elements can be the output from a deep CNN, but results in a dense affinity matrix that effectively models any task-specific pairwise similarity matrix. The spatial propagation network can be applied to many affinity-related tasks, such as image matting, segmentation and colorization, to name a few. Essentially, the model can learn semantically aware affinity relations for high-level vision tasks due to the powerful learning capability of the deep CNN. We validate the framework on the task of refinement for image segmentation boundaries. Experiments on face parsing and semantic segmentation tasks show that the spatial propagation network provides a general, effective, and efficient solution for generating high-quality segmentation results.  Back
 
Topics:
AI and DL Research, Computer Vision, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8312
Streaming:
Download:
Share: