April 20, 2016 at 23:52Spring meeting of Pattern Recognition and Computer Vision Colloquium 2016 in Prague: how it was
Good evening, habrasociety!
I already wrote about the summer school of computer vision, which was held last summer at the Czech Technical University in Prague in an article . At the end of March, I was lucky to take part in another one-day event of this kind, organized by the same group The Center for Machine Perception from the CTU in Prague. This time it was a series of 6 reports in breaks for lunch and a coffee break. The main topics of the event were algorithms for computer vision and machine learning. Judging by the large number of reports on Deep learning, this direction is becoming a very popular trend in solving computer vision problems. I ask for details under cat. Caution traffic!
The event took place in a very beautiful historical building of the Faculty of Electrical Engineering in the center of Prague.
As lecturers, as in the summer school, there were invited scientific experts from various universities from around the world engaged in research in the field of computer vision. Among them was also a doctoral student from CTU.
Half of the lectures were devoted to machine learning, including graphical models, deep learning, and feature maps.
Antonin Sulc spoke about an interesting direction in computer vision - light fields and its practical application. To begin with, he introduced the technology itself, which arose thanks to the development of compact light-field cameras Lytro Illum. In addition to color information, such cameras also provide information on the angle of incidence of rays in each pixel. The simplest application of such cameras is to change the focus of the camera from the background to objects in the foreground and vice versa. The light-field technology is used in structure-from motion, it allows you to split the image into several layers (for example, highlight reflective and transparent surfaces in the image) and effectively evaluate the depth (build a depth map).
Doctoral student Giorgos Tolias from CTU talked about his development of a feature descriptor that allows you to search for images using a sketch made by hand as a request.
The most interesting in my opinion was the report by Yannis Avrizisa, which talked about approximate search methods for solving clustering problems in multidimensional space. He described the problem of searching in multidimensional space, and showed in which computer vision problems this problem is being solved (image search, estimation of the shooting position, Bag of words and clustering). Various methods of approximate search were also presented such as binary codes, quantization (vector, product quantization) and clustering (k-means and its various modifications, Gaussian mixtures ranked retrieval and inverted-quantized k-means). All methods were presented with a detailed mathematical explanation and shown graphically in graphical form.
You can familiarize yourself with the whole program and download presentations at the link:cmp.felk.cvut.cz/cmp/events/colloquium-2016.03.31 .
After the main part was over, everyone made a short excursion to the workrooms of the CMP center.
In the beginning we were shown an online search engine for similar images for a given image fragment. The user can frame the object of interest (for example, the dome of the cathedral) on the image and in response receives all relevant images that may contain this object from the same perspective, from different viewpoints, at different scales and even more detailed.
Here is a link to an article about the project for those who want to know more about the project: cmp.felk.cvut.cz/~chum/papers/mikulik_sisap13.pdf .
As a result, the day turned out to be eventful and informative. I hope that in the future it will be possible to attend other similar events. This event - Pattern Recognition and Computer Vision Colloquium - is held every six months, the last time it was held in November 2015. Such events allow me to get acquainted with experts and learn about new trends and technologies in the field of computer vision.
Good luck to everyone and wait for new reviews!
I already wrote about the summer school of computer vision, which was held last summer at the Czech Technical University in Prague in an article . At the end of March, I was lucky to take part in another one-day event of this kind, organized by the same group The Center for Machine Perception from the CTU in Prague. This time it was a series of 6 reports in breaks for lunch and a coffee break. The main topics of the event were algorithms for computer vision and machine learning. Judging by the large number of reports on Deep learning, this direction is becoming a very popular trend in solving computer vision problems. I ask for details under cat. Caution traffic!
The event took place in a very beautiful historical building of the Faculty of Electrical Engineering in the center of Prague.
As lecturers, as in the summer school, there were invited scientific experts from various universities from around the world engaged in research in the field of computer vision. Among them was also a doctoral student from CTU.
Half of the lectures were devoted to machine learning, including graphical models, deep learning, and feature maps.
Antonin Sulc spoke about an interesting direction in computer vision - light fields and its practical application. To begin with, he introduced the technology itself, which arose thanks to the development of compact light-field cameras Lytro Illum. In addition to color information, such cameras also provide information on the angle of incidence of rays in each pixel. The simplest application of such cameras is to change the focus of the camera from the background to objects in the foreground and vice versa. The light-field technology is used in structure-from motion, it allows you to split the image into several layers (for example, highlight reflective and transparent surfaces in the image) and effectively evaluate the depth (build a depth map).
Doctoral student Giorgos Tolias from CTU talked about his development of a feature descriptor that allows you to search for images using a sketch made by hand as a request.
The most interesting in my opinion was the report by Yannis Avrizisa, which talked about approximate search methods for solving clustering problems in multidimensional space. He described the problem of searching in multidimensional space, and showed in which computer vision problems this problem is being solved (image search, estimation of the shooting position, Bag of words and clustering). Various methods of approximate search were also presented such as binary codes, quantization (vector, product quantization) and clustering (k-means and its various modifications, Gaussian mixtures ranked retrieval and inverted-quantized k-means). All methods were presented with a detailed mathematical explanation and shown graphically in graphical form.
You can familiarize yourself with the whole program and download presentations at the link:cmp.felk.cvut.cz/cmp/events/colloquium-2016.03.31 .
After the main part was over, everyone made a short excursion to the workrooms of the CMP center.
In the beginning we were shown an online search engine for similar images for a given image fragment. The user can frame the object of interest (for example, the dome of the cathedral) on the image and in response receives all relevant images that may contain this object from the same perspective, from different viewpoints, at different scales and even more detailed.
Here is a link to an article about the project for those who want to know more about the project: cmp.felk.cvut.cz/~chum/papers/mikulik_sisap13.pdf .
As a result, the day turned out to be eventful and informative. I hope that in the future it will be possible to attend other similar events. This event - Pattern Recognition and Computer Vision Colloquium - is held every six months, the last time it was held in November 2015. Such events allow me to get acquainted with experts and learn about new trends and technologies in the field of computer vision.
Good luck to everyone and wait for new reviews!