Search

This paper investigates modifying an existing neural network architecture for static saliency prediction using two types of recurrences that integrate information from the temporal domain. The first modification is the addition of a ConvLSTM within the architecture, while the second is a conceptually simple exponential moving average of an internal convolutional state. We use weights pre-trained on the SALICON dataset and fine-tune our model on DHF1K. Our results show that both modifications achieve state-of-the-art results and produce similar saliency maps. Source code is available at https://git.io/fjPiB., Accepted at BMVC 2019

Insight-DCU participated in the instance search (INS), semantic indexing (SIN), and localization tasks (LOC) this year. In the INS task we used deep convolutional network features trained on external data and the query data for this year to train our system. We submitted four runs, three based on convolutional network features, and one based on SIFT/BoW. F A insightdcu 1 was an automatic run using features from the last convolutional layer of a deep network with bag-of-words encoding and achieved 0.123 mAP. F A insightdcu 2 modied the previous run to use re-ranking based on an R-CNN model and achieved 0.111 mAP. I A insightdcu 3, our interactive run, achieved 0.269 mAP. Our SIFT-based run F A insightdcu 2 used weak geometric consistency to improve performance over the previous year to 0.187 mAP. Overall we found that using features from the convolutional layers improved performance over features from the fully connected layers used in previous years, and that weak geometric consistency improves performance for local feature ranking. In the SIN task we again used convolutional network features, this time netuning a network pretrained on external data for the task. We submitted four runs, 2C D A insightdcu.15 1..4 varying the top-level learning algorithm and use of concept co-occurance. 2C D A insightdcu.15 1 used a linear SVM top-level learner, and achieved 0.63 mAP. Exploiting concept co-occurance improved the accuracy of our logistic regression run 2C D A insightdcu.15 3 from 0.058 mAP to 0.6 2C D A insightdcu.15 3. Our LOC system used training data from IACC.1.B and features similar to our INS run, but using a VLAD encoding instead of a bag-of-words. Unfortunately there was problem with the run that we are still investigating.

Insight-DCU participated in the instance search (INS) and semantic indexing (SIN) tasks in 2014. Two very different approaches were submitted for instance search, one based on features extracted using pre-trained deep convolutional neural networks (CNNs), and another based on local SIFT features, large vocabulary visual bag-of-words aggregation, inverted index-based lookup, and geometric verification on the top-N retrieved results. Two interactive runs and two automatic runs were submitted, the best interactive runs achieved a mAP of 0.135 and the best automatic 0.12. Our semantic indexing runs were based also on using convolutional neural network features, and on Support Vector Machine classifiers with linear and RBF kernels. One run was submitted to the main task, two to the no annotation task, and one to the progress task. Data for the no-annotation task was gathered from Google Images and ImageNet. The main task run has achieved a mAP of 0.086, the best no-annotation runs had a close performance to the main run by achieving a mAP of 0.080, while the progress run had 0.043.

Dublin City University participated with a consortium of colleagues from NUI Galway and Universitat Politecnica de Catalunya in two tasks in TRECVid 2016, Instance Search (INS) and Video to Text (VTT). For the INS task we developed a framework consisting of face detection and representation and place detection and representation, with a user annotation of top-ranked videos. For the VTT task we ran 1,000 concept detectors from the VGG-16 deep CNN on 10 keyframes per video and submitted 4 runs for caption re-ranking, based on BM25, Fusion, word2vec and a fusion of baseline BM25 and word2vec. With the same pre-processing for caption generation we used an open source image-to-caption CNN-RNN toolkit NeuralTalk2 to generate a caption for each keyframe and combine them.

This paper outlines 6 approaches taken to computing video memorability, for the MediaEval media memorability task. The approaches are based on video features, an end-to-end approach, saliency, aesthetics, neural feedback, and an ensemble of all approaches.