Near-Duplicate Video Retrieval with Deep Metric Learning

Abstract

This work addresses the problem of Near-Duplicate Video Retrieval (NDVR). We propose an effective videolevel NDVR scheme based on deep metric learning that leverages Convolutional Neural Network (CNN) features from intermediate layers to generate discriminative global video representations in tandem with a Deep Metric Learning (DML) framework with two fusion variations, trained to approximate an embedding function for accurate distance calculation between two near-duplicate videos. In contrast to most state-of-the-art methods, which exploit information deriving from the same source of data for both development and evaluation (which usually results to dataset-specific solutions), the proposed model is fed during training with sampled triplets generated from an independent dataset and is thoroughly tested on the widely used CC WEB VIDEO dataset, using two popular deep CNN architectures (AlexNet, GoogleNet). We demonstrate that the proposed approach achieves outstanding performance against the state-of-the-art, either with or without access to the evaluation dataset.

Verification of Web Videos Through Analysis of Their Online Context

“…• The video title, event title, and the four translations are submitted as search queries to the three target platforms (YouTube, Facebook, Twitter) and all results are aggregated in a common pool. • Using the near-duplicate retrieval algorithm of Kordopatis-Zilos et al [27], we filter the pool of videos in order to discard unrelated ones. • Finally a manual confirmation step is used to remove erroneous results of the automatic method and only retain actual near-duplicates.…”

“…A viral video of a girl being chased by a bear while snowboarding was posted five days after the channel was created. The video gained millions of views before it was debunked 27 . The average number of videos per month uploaded by the channel is calculated by dividing the total number of videos posted by the channel to the number of months that this channel is alive.…”

“…eral cases of debunked videos along with their near-duplicates, accompanied with comments about the false claim, a label indicating the type of the manipulation and optionally a link to an article or a video debunking it. The near-duplicate algorithm of [27] is used to search within the pool of already debunked videos of the FVC-2018. If there is a match, then CAA checks whether the video is unique or has nearduplicates.…”

Finding Near-Duplicate Videos in Large-Scale Collections

“…To observe no performance loss, a sufficiently large and diverse dataset to create vocabularies is required, which needs significant effort to be collected or sometimes is not even possible. Hence, we also developed a Deep Metric Learning (DML) approach to overcome this limitation [28]. This involves training a Deep Neural Network (DNN) to approximate an embedding function for the accurate computation of similarity between two candidate videos.…”

“…In Section 4.2, we review the related literature in the field of NDVR by providing an outline of the major trends in the field. In Section 4.3, we present the two aforementioned NDVR approaches that have been developed within the InVID project [27,28]. In Section 4.4, we report on the results of a comprehensive experimental study, including a comparison with five state-of-the-art methods.…”