Weakly-Supervised Dense Video Captioning (WSDVC) aims to localize and describe all events of interest in a video without requiring annotations of event boundaries. This setting poses a great challenge in accurately locating the temporal location of event, as the relevant supervision is unavailable. Existing methods rely on explicit alignment constraints between event locations and captions, which involve complex event proposal procedures during both training and inference. To tackle this problem, we propose a novel implicit location-caption alignment paradigm by complementary masking, which simplifies the complex event proposal and localization process while maintaining effectiveness. Specifically, our model comprises two components: a dual-mode video captioning module and a mask generation module. The dual-mode video captioning module captures global event information and generates descriptive captions, while the mask generation module generates differentiable positive and negative masks for localizing the events. These masks enable the implicit alignment of event locations and captions by ensuring that captions generated from positively and negatively masked videos are complementary, thereby forming a complete video description. In this way, even under weak supervision, the event location and event caption can be aligned implicitly. Extensive experiments on the public datasets demonstrate that our method outperforms existing weakly-supervised methods and achieves competitive results compared to fully-supervised methods.

More recently, developing 2D and 3D convolutional neural networks (CNNs) has sparked interest in studying static and dynamic visual media's encoding, captioning, and query-answering capabilities. However, accomplishing these tasks on long, unedited video significantly challenges computer vision. Dense video captioning aims to make a computer understand what is happening in a video and establish a relation between the video content and its meaningful natural language description. The capability of describing events in videos aids a variety of systems, including blind navigation, video searching, surveillance, medical image analysis, and automatic video subtitling. The urge to detect captions on images and videos started in 1970 when researchers began working with images and video snippets containing captions. The art of displaying text on images and video transcribing the audio is called closed captioning. To serve the consumers who are hard of hearing and to take part in technology improvement motivated researchers to develop some automatic caption detection systems [92, 152].

In this work, we introduce Vid2Seq, a multi-modal single-stage dense event captioning model pretrained on narrated videos which are readily-available at scale. The Vid2Seq architecture augments a language model with special time tokens, allowing it to seamlessly predict event boundaries and textual descriptions in the same output sequence. Such a unified model requires large-scale training data, which is not available in current annotated datasets. We show that it is possible to leverage unlabeled narrated videos for dense video captioning, by reformulating sentence boundaries of transcribed speech as pseudo event boundaries, and using the transcribed speech sentences as pseudo event captions. The resulting Vid2Seq model pretrained on the YT-Temporal-1B dataset improves the state of the art on a variety of dense video captioning benchmarks including YouCook2, ViTT and ActivityNet Captions. Vid2Seq also generalizes well to the tasks of video paragraph captioning and video clip captioning, and to few-shot settings. Our code is publicly available at https://antoyang.github.io/vid2seq.html.

Detecting suspicious activities in surveillance videos is a longstanding problem in real-time surveillance that leads to difficulties in detecting crimes. Hence, we propose a novel approach for detecting and summarizing suspicious activities in surveillance videos. We have also created ground truth summaries for the UCF-Crime video dataset. We modify a pre-existing approach for this task by leveraging the Human-Object Interaction (HOI) model for the Visual features in the Bi-Modal Transformer. Further, we validate our approach against the existing state-of-the-art algorithms for the Dense Video Captioning task for the ActivityNet Captions dataset. We observe that this formulation for Dense Captioning performs significantly better than other discussed BMT-based approaches for BLEU@1, BLEU@2, BLEU@3, BLEU@4, and METEOR. We further perform a comparative analysis of the dataset and the model to report the findings based on different NMS thresholds (searched using Genetic Algorithms). Here, our formulation outperforms all the models for BLEU@1, BLEU@2, BLEU@3, and most models for BLEU@4 and METEOR falling short of only ADV-INF Global by 25% and 0.5%, respectively.