Fine-grained and instance-level recognition methods are commonly trained and evaluated on specific domains, in a model per domain scenario. Such an approach, however, is impractical in real large-scale applications. In this work, we address the problem of universal image embedding, where a single universal model is trained and used in multiple domains. First, we leverage existing domain-specific datasets to carefully construct a new large-scale public benchmark for the evaluation of universal image embeddings, with 241k query images, 1.4M index images and 2.8M training images across 8 different domains and 349k classes. We define suitable metrics, training and evaluation protocols to foster future research in this area. Second, we provide a comprehensive experimental evaluation on the new dataset, demonstrating that existing approaches and simplistic extensions lead to worse performance than an assembly of models trained for each domain separately. Finally, we conducted a public research competition on this topic, leveraging industrial datasets, which attracted the participation of more than 1k teams world-wide. This exercise generated many interesting research ideas and findings which we present in detail. Project webpage: https://cmp.felk.cvut.cz/univ_emb/

Understanding video content and generating caption with context is an important and challenging task. Unlike prior methods that typically attempt to generate generic video captions without context, our architecture contextualizes captioning by infusing extracted information from relevant text data. We propose an end-to-end sequence-to-sequence model which generates video captions based on visual input, and mines relevant knowledge such as names and locations from contextual text. In contrast to previous approaches, we do not preprocess the text further, and let the model directly learn to attend over it. Guided by the visual input, the model is able to copy words from the contextual text via a pointer-generator network, allowing to produce more specific video captions. We show competitive performance on the News Video Dataset and, through ablation studies, validate the efficacy of contextual video captioning as well as individual design choices in our model architecture.

We propose an end-to-end approach for phrase grounding in images. Unlike prior methods that typically attempt to ground each phrase independently by building an image-text embedding, our architecture formulates grounding of multiple phrases as a sequential and contextual process. Specifically, we encode region proposals and all phrases into two stacks of LSTM cells, along with so-far grounded phrase-region pairs. These LSTM stacks collectively capture context for grounding of the next phrase. The resulting architecture, which we call SeqGROUND, supports many-to-many matching by allowing an image region to be matched to multiple phrases and vice versa. We show competitive performance on the Flickr30K benchmark dataset and, through ablation studies, validate the efficacy of sequential grounding as well as individual design choices in our model architecture.

Abstract In a photo, motion blur can be used as an artistic style to convey motion and to direct attention. In panning or tracking shots, a moving object of interest is followed by the camera during a relatively long exposure. The goal is to get a blurred background while keeping the object sharp. Unfortunately, it can be difficult to impossible to precisely follow the object. Often, many attempts or specialized physical setups are needed. This paper presents a novel approach to create such images. For capturing, the user is only required to take a casually recorded hand‐held video that roughly follows the object. Our algorithm then produces a single image which simulates a stabilized long time exposure. This is achieved by first warping all frames such that the object of interest is aligned to a reference frame. Then, optical flow based frame interpolation is used to reduce ghosting artifacts from temporal undersampling. Finally, the frames are averaged to create the result. As our method avoids segmentation and requires little to no user interaction, even challenging sequences can be processed successfully. In addition, artistic control is available in a number of ways. The effect can also be applied to create videos with an exaggerated motion blur. Results are compared with previous methods and ground truth simulations. The effectiveness of our method is demonstrated by applying it to hundreds of datasets. The most interesting results are shown in the paper and in the supplemental material.

The alignment of heterogeneous sequential data (video to text) is an important and challenging problem. Standard techniques for this task, including Dynamic Time Warping (DTW) and Conditional Random Fields (CRFs), suffer from inherent drawbacks. Mainly, the Markov assumption implies that, given the immediate past, future alignment decisions are independent of further history. The separation between similarity computation and alignment decision also prevents end-to-end training. In this paper, we propose an end-to-end neural architecture where alignment actions are implemented as moving data between stacks of Long Short-term Memory (LSTM) blocks. This flexible architecture supports a large variety of alignment tasks, including one-to-one, one-to-many, skipping unmatched elements, and (with extensions) non-monotonic alignment. Extensive experiments on semi-synthetic and real datasets show that our algorithm outperforms state-of-the-art baselines.