-
Notifications
You must be signed in to change notification settings - Fork 0
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
News item detection in RBB content #4
Comments
Let's start by determining the different classes of temporal fragments in an RBB news show (i.e., intro, close, weather, kiosk (newspapers), interview, etc.). So, it would be a good start if the RBB partners prepare a list with all these different categories. Afterwards, we have to find out which one is the most appropriate way to detect these fragments and to assign the needed label (according to the provided list of classes/labels) to each one of them. In general, this is something duable using a combination of 3 different visual analysis techniques (i.e. shot segmentation, concept detection and bumper re-detection) and a new set of rules for finding the appropriate cuts between the different chapters, but it's really hard to have it fully tested and running before August 15. Regarding the last question, I believe that the integration one more label/attribute for each defined chapters, in the Exmaralda is not a big trouble (but I have to discuss this with DanielS). |
Next step @nicorbb "prepare a list with all these different categories" |
When do you expect these inputs and - actually - why?
IMHO, RBB's editors do not expect more than this! There is no request (that I have heard of) to automatically detect what kind of content can be found inside a chapter! We only want clear start and end marks for a chapter and for that the existing strategies seem to be sufficient!? |
@nicorbb yes of course we can leave it to editors to tag an automatically extracted video chapter as being a News Item (see also Editor Tool issue linkedtv/editortool#15) |
From CERTH's side, as I already mentioned, the annotation of different detected parts of the video is something duable, after defining the different categories of chapters and integrating a set of rules in the current software. However, I would say that a bilateral discussion between CERTH and RBB would be the best approach for concluding if the current algorithm is sufficient enough or something more is needed. So, let's say that this issue is closed for now, but we can re-open it again if a new demand from RBB's side appears. |
Yes. And no ;-) I feel the need for action is on me now, but I am still not sure what exactly is expected and to which end. Could we just have a small talk on this next week in St Gallen? |
Yes, I think that a small talk (and maybe a brief video demo) at St. Gallen, in order to clarify the analysis capability of the current RBB chapter segmentation algorithm and find potential extensions of it, is the best that we can do. |
I reopen this. So, my understanding is RBB communicated that based on CERTHs current news bumper detection you can consider every chapter excluding the first and last chapter in a news video to be a NEWS ITEM. Correct? |
"So, my understanding is RBB communicated that based on CERTHs current news bumper detection you can consider every chapter excluding the first and last chapter in a news video to be a NEWS ITEM. Correct?" Yes! "If so, I close this issue once WP1 confirms it can automatically label every RBB chapter according to this definition as NEWS ITEM in the REST service used for the LinkedNews scenario (configuration). AFAIK you need to (ad hoc) define news items inside the EXB file in some way (IMHO it doesnt matter how exactly as long as you document this and communicate it to EURECOM for TV2RDF, i.e. by adding a comment to issue" I can confirm that CERTH plans to do this. To be more precise, this is something implemented (yesterday) and will be supported by the new version of the WP1 REST Service (you will receive a detailed e-mail about the updated WP1 REST Service, in a few moments). Jose-Luis will be informed appropriately about the changes in the EXMARaLDA file and then I will post a comment to issue linkedtv/ontologies#8 as you suggested. |
Can WP1 "detect" using the bumper detection for chapters and an assumption on the common structure of every RBB news story which chapters are indeed "news items" as opposed to intros, closes, or other news segments (weather, interview etc.)
Can this be tagged somehow in the Exmaralda?
The text was updated successfully, but these errors were encountered: