Identifying visual entities in a collection of INA videos or in your images
Find DigInPix in INA's Research Area.
DigInPix is a system that automatically recognizes visual entities coming from dictionaries in INA videos or in your images. The main purpose of putting this prototype online is to demonstrate the performance of our algorithms and prove the benefits of such a system in helping to annotate multimedia content like INA's. This application features state-of-the-art algorithms and has the distinctive feature of being able to work on large volumes of data to detect very small objects in images.
The visual dictionaries contain a series of visual entities: logos, car models, pictures, buildings, etc. In this demonstration you can access a dictionary of 25,000 logos, as well as a dictionary of 25 000 paintings from the Images d'Art website of Réunion des musées nationaux - Grand Palais (Rmn-GP). Each entity (logo or painting) is represented by one or several images in the dictionary to enable DigInPix to learn to recognize it.
We offer three collections from INA's archives: 1,000 videos of sport events, 650 videos of art and 1,480 videos from Ina.fr. You can also test searches in your own images on your hard drive or in images on the web.
Search in DigInPix
Search with DigInPix
There are several access options for using DigInPix with INA's collections:
- Text field: to search by name for entities in the dictionaries and documents in the collections. The results are displayed in two different tabs.
- Browse our Collections data: to discover all videos with the number of entities detected in each one.
- Browse our Dictionaries data: to discover all entities in the dictionary. For each entity you can see all the videos in which it has been detected or all the images that enabled the DigInPix system to learn to define the entity.
Search in DigInPix
View the results
The results document can be sorted by title, or by the number of visual entities detected in each document, or even by the relevance rating of the text query. You can also sort the entity results, in which case that is the number of documents containing this entity.
You can view the results of each video by clicking it. The results of video analysis by the DigInPix system are presented on a dedicated page.
You can use an enhanced player to watch the video and locate on the timelines the visual entities detected by DigInPix. For each entity, the overall confidence rating and the scores for each detected item are shown. Click an entity to see the result on the corresponding keyframe of the video. You can also easily browse through each detected entity using the previous and next buttons on each of the lines.
Under the player, detected entities are displayed with their score and a representative image.
Search in DigInPix
Analysis of your content
To test DigInPix, you can send your photos and observe the results:
- either by pasting the URL of an image found on the web
- or by uploading an image from your computer
Naturally, detected entities are limited to those in the visual dictionary, in other words in this demonstration of logos. As for the videos, the performance of DigInPix on your images depends on the criteria referred to earlier.
Make sure you comply with the GTU when using this demonstrator, more particularly with regard to the intellectual property of content and rights of personal portrayal.
The analysis result of your images is accessible via an URL we provide you with that you alone know. You can share the URL to show detected entities. In which case your images will become public.
The performance of DigInPix varies according to the following criteria:
- visual features of the entity
- richness of representative images in the dictionaries
- quality and size of analyzed documents
- size of objects sought in these documents
Search in DigInPix
Your remarks
DigInPix is a prototype being developed by INA's Research department. It is currently a beta version. As such, we would appreciate any opinions and remarks you may want to share with us. Whether you are enthusiastic or disappointed with the results, whether you wish to work with us or merely find out more, contact us!
Search in DigInPix