| Channel: cvpr2011 |
Your post is to long. Please make it short.
new posts available!
| wow just gone to "object detection" session, the hall is overfilled :) seems to be that this is the most favourite topic. Unfortunately the talk I want to visit is not about meshes although the title made this hope... | |||
|
| hmm strange, they showed this actually already as "cooperative cuts" in a workshop at monday... The contribution is to change edge cost function adopting them to more global information,like edges helps other edges to improve on their costs. | |||
|
| next: Submodularity beyound submodular energies: coupling edges in graph cuts from our collegeus at MPI for Intelligent Systems :) | |||
|
| Improvement on an optimization technique. The main contribution is an optimized Nesterov's optimization scheme, for example a better stopping condition. I need to know more about Nesterov's method to provide more confident comment about this work... | |||
|
| next: A Study of Nesterov's Scheme for Lagrangian Decomposition and MAP Labeling. | |||
|
| honestly I didn't quite get the idea. Seems to be another pairwise potential function for better solving of specific optimization problem. Nobody asked any question, so either nobody also realy get this or everything was clear and I am just stupid :( | |||
|
| New day and I found myself in the "Optimization Methods" session. First talk: A Non-Convex Relaxation Approach to Sparse Dictionary Learning. A speaker speaks almost without breaks between the words and with constant intonation. Hard to get that :) | |||
|
| New day and I found myself in the "Optimization Methods" session. First talk: A Non-Convex Relaxation Approach to Sparse Dictionary Learning. A speaker speaks almost without breaks between the words and with constant intonation. Hard to get that :) | |||
|
| "Discrete-continuous Optimization for Large Scale Structure from Motion" is really cool one. The authors presented a loopy belief propagation method to optimize bundle adjustment in scene reconstruction. Their labelset is also really huge as in our work | |||
|
| next:Single-image shadow detection and removal using paired regions.Authors detect shadow regions by a energy minimization approach where a pair of clustered parts is considered,i.e. parts of different illumination or same ill. Hard shadow map is a result | |||
|
| next talk: a generative model for 3d urban scene understanding from movable platforms. The idea is to reconstruct parameters of roads and intersections from a video. The approach can estimate the way how a car should handle automatically on intersections. | |||
|
| really cool paper. the authors look for human doable tasks in the scene,e.g. where human can sit in the image. This work is so exciting,that more questions arises then it has benn explained in the talk. | |||
|
| really cool paper. the authors look for human doable tasks in the scene,e.g. where human can sit in the image. This work is so exciting,that more questions arises then it has benn explained in the talk. | |||
|
| found my way to scene understanding session. first talk: From 3d scene geometry to human workspaces | |||
|
| so they propose different optics and design tools to build this devices. optics solves a part of cv problems. these device can then be used my microbots which are couple of mm in size. pretty cool! | |||
|
| next: Wide_angle micro sensors for vision on a tight budget. Pretty cool, they use the idea presented couple of years ago, to employ optics for solving computer vision tasks. | |||
|
| next talk by Ilja about the kaleidoscopic imaging. there were no tought question for him :) congrats to Ilja for his first international conference oral presentation ! | |||
|
| the main work doje by authors is to compute high resolution motion vectors from low res video. Then using this information they reconstruct the actual video. | |||
|
| very bad wifi connection here. Paper "A bayesian approach to adaptive video super resolution" is considering to look at. Very impressive results on video super resolution. | |||
|
| first impression: it is very crowdy. First poster session was particularly not enjoyable because of so much people in so small rooms :( | |||
|