September 28, 2011

Seminar Announcement (Sep 29, 2011)

Filed under: seminar — scyscyao @ 10:16 am
The seminar this week is scheduled on Thursday, Sep.29 at 10 am.
The paper I will be presenting is:
“Birdlets: Subordinate Categorization Using Volumetric Primitives and Pose-Normalized Appearance”, ICCV11.

This paper proposes a framework for birds detection and subordinate categorization. For birds detection, the authors employ a framework based on Poselets to detect volumetric part model of birds. For subordinate categorization, the authors employ Pose-Normalized Appearance Descriptor(PNAD) coupled with the Stacked Evidence Tree to categorize birds of different family subset. 

Main references for the tools used in the paper is:
1. Poselets: [7] [8], slides for ICCV 09 paper on the website: http://www.eecs.berkeley.edu/~lbourdev/poselets/poselets.pptx (Recommended to take a look in “Slide Show” model before attending the seminar)

2. PNAD: [13] [33], website for color-SIFT: http://staff.science.uva.nl/~mark/downloads.html
3. Stacked Evidence Tree: [39] [10]
Slides can be downloaded here

September 21, 2011

Seminar Announcement (Sep 22, 2011)

Filed under: seminar — scyscyao @ 7:39 pm

(adapted from Shahzor’s E-mail)
The seminar for this week is scheduled on Thursday, 22nd at 10 am.

I will be presenting the following PAMI2010 paper:
X. Liu, O. Veksler, and J. Samarabandu, “Order-preserving moves for graph-cut based optimizaton“, PAMI 2010.

In this paper, ordering constraints are incorporated in the Potts model for the smoothness energy function. However, they show that the alpha-expansion move algorithm commonly used in graph cuts is trapped in a local minimum when using ordering constraints. They develop two new ‘order preserving’ moves, that can be used with ordering constraints and yield the globally optimal results. However, the technique is model-specific and applied specifically to the ‘five party model’.

I will also try to touch on the following, which also uses ordering constraints:
P. F. Felzenszwalb, O. Veksler, “Tiered Scene Labeling with Dynamic Programming.”, CVPR 2010.


Update: The slides can be downloaded here

Slides for Yuxiang’s presentation last week

Filed under: seminar — scyscyao @ 4:49 pm
The slides can be downloaded here
Here is Yuxiang’s announcement extracted from his email:
The paper I will be presenting this afternoon is:
Efficient Computation of Robust Low-Rank Matrix Approximations in the Presence of Missing Data using the L1 Norm, cvpr 2010, best paper award.

The paper and an important one-page ERRATA can be found in the author’s website: http://cs.adelaide.edu.au/~anders/

Without this errata, it will be difficult to follow the derivations and ideas in the paper. (they are also attached for your convenience)

Related papers are:

Original Wiberg: [16] Takayuki Okatani and Koichiro Deguchi,On the Wiberg Algorithm for Matrix Factorization in the Presence of Missing Components

ALP and AQP: [13] Ke and Kanade, Robust L1 Norm Factorization in the Presence of Outliers and Missing Data by Alternative Convex Programming

June 27, 2011

Seminar Announcement (July 01 2011)

Filed under: seminar — Tags: , , — Ju Sun @ 5:23 pm

For the forthcoming seminar, I’m going to discuss about the state-of-the-art on subspace segmentation and estimation (SSE) powered by the recent progress on sparsity and low rank optimization (we’ve gained some partial impression on this before);  Meanwhile I’d like to highlight some open problems in this avenue, without solving which we cannot claim sensible progress on the SSE problem despite the emerging promise.
Title: Optimization-based subspace segmentation and open problems
Time : 2 — 4pm, Friday (July 1st 2011)  
Venue: Meeting room in Vision and Machine Learning Lab (E4 – 08 -27)

(from Sun Ju)

Dr. gao’s list of References (for the passing Seminar)

Filed under: reference, seminar — Tags: , , — sfxnus @ 5:20 pm

(adapted from Dr. Gao’s email)

May 27, 2011

Slides for “A Model of Bee Vision”

Filed under: general reading — Tags: , , — scyscyao @ 11:09 pm

Hi all,

The slides of “A model of Bee Vision” can be downloaded at google docs or dropbox

The main reference of the presentation is the book “What does the Honeybee See? And How do We Know?” which is published in 2009. It can be downloaded at: http://epress.anu.edu.au/honeybee/pdf/whole_book.pdf

Despite its low resolution and poor ability in shape and pattern recognition, bee’s vision system is proved to be an effective system for places recognition and navigation. It seems that the bee memorizes a whole panaroma as a “pattern” and use this “pattern” the recognize a place. And its local region (the smallest area in which same pattern can be discriminated, 25-30 degree) can easily discriminate flow-like shape (radial or tangential, though it can not discriminate other shapes like triangle or rectangle). There are also behavioral evidence that the bee utilizes the optic flow to avoid obstacle and collision when flying and to ensure smooth landing on a surface.

The model mentioned in the presentation is mainly on the pattern recognition ability of bee. The model mush have limitations, but from it, as least now, we could conclude that an effective vision system do not necessarily possess the resolution and processing ability of human vision. And isolating the vision system from the whole system might not be a good idea to developing a powerful artificial vision system.

May 25, 2011

Optional readings for Choon Meng’s seminar and some preliminary Comments

Filed under: seminar — Tags: , — Ju Sun @ 1:14 pm

In addition to the main reference, I’d like to recommend two additional readings as complementary material

Besides, there’s one caution on reading the main reference. In Lemma 1 – 3, the authors have used the term “the optimal solution” without proper justification on the uniqueness (hence “the” instead of “an”). In other words, the existence of one minimizer does not imply there are no other minimizers. On the other hand, the original proofs are most by construction, which are hard to see and have little use for generalization. I’m writing a blog article in an attempt to handle the two problems (in progress, hopefully finished by this weekend)



Filed under: Uncategorized — leechoonmeng @ 10:43 am

The paper for this coming seminar discussion is

A Closed Form Solution to Robust Subspace Estimation and Clustering (PDF)
Paolo Favaro, RenéVidal, Avinash Ravichandran (Johns Hopkins University)

Date: Wednesday, 1st June 2011
Time: 2 – 4 pm
Venue: E4-04-04

May 10, 2011

Two general readings on research Methodology

Filed under: general reading — Tags: , , — sfxnus @ 3:55 pm

All kinds of things you may be thinking of in research: motivation, luck, intellectual ability, age (fame and early success), working conditions, commitment and passion, courage, open-mindedness, selling your work; and also possible impediments to success — I suggest you try to summarize each point while reading, as it contains a lot of pieces.

Brook’s view of what we can expect from intelligent machines — robots. Do we need just replicate what we are as humans into machines, or do we make specialized machines that do particular tasks well — even better than what we can (hyper-intelligence)?

January 24, 2011

Slides for “Rigid Structure from Motion from a Blind Source Separation Perspective” presentation

Filed under: reference — leechoonmeng @ 3:44 pm

The slides for today’s presentation can be found here

Older Posts »

Blog at WordPress.com.