Grounding Object Detections With Transcriptions
This work addresses the challenge of expensive and error-prone human annotation for video data, offering an incremental step towards automating training set construction for specific tasks.
The paper tackles the problem of lacking labeled data for supervised models in audio-visual content by proposing a method to automatically extract entity-video frame pairs from instruction videos using speech transcriptions and videos, resulting in experiments on image recognition and visual grounding tasks with evaluation on manually annotated datasets.
A vast amount of audio-visual data is available on the Internet thanks to video streaming services, to which users upload their content. However, there are difficulties in exploiting available data for supervised statistical models due to the lack of labels. Unfortunately, generating labels for such amount of data through human annotation can be expensive, time-consuming and prone to annotation errors. In this paper, we propose a method to automatically extract entity-video frame pairs from a collection of instruction videos by using speech transcriptions and videos. We conduct experiments on image recognition and visual grounding tasks on the automatically constructed entity-video frame dataset of How2. The models will be evaluated on new manually annotated portion of How2 dev5 and val set and on the Flickr30k dataset. This work constitutes a first step towards meta-algorithms capable of automatically construct task-specific training sets.