Compositional image retrieval (CIR) is a multimodal learning task where a model combines a query image with a user-provided text modification to retrieve a target image. CIR finds applications in a variety of domains including product retrieval (e-commerce) and web search. Existing methods primarily focus on fully-supervised learning, wherein models are trained on datasets of labeled triplets such as FashionIQ and CIRR. This poses two significant challenges: (i) curating such triplet datasets is labor intensive; and (ii) models lack generalization to un-seen objects and domains. In this work, we propose SCOT (Self-supervised COmpositional Training), a novel zero-shot compositional pretraining strategy that combines existing large image-text pair datasets with the generative capabilities of large language models to contrastively train an embedding composition network. Specifically, we show that the text embedding from a large-scale contrastively-pretrained vision-language model can be utilized as proxy target supervision during compositional pretraining, replacing the target image embedding. In zero-shot settings, this strategy surpasses SOTA zero-shot compositional re-trieval methods as well as many fully-supervised methods on standard benchmarks such as FashionIQ and CIRR. Our code and models are available at https://github.com/yahoo/SCOT.

We present a model for temporally precise action spotting in videos, which uses a dense set of detection anchors, predicting a detection confidence and corresponding fine-grained temporal displacement for each anchor. We experiment with two trunk architectures, both of which are able to incorporate large temporal contexts while preserving the smaller-scale features required for precise localization: a one-dimensional version of a u-net, and a Transformer encoder (TE). We also suggest best practices for training models of this kind, by applying Sharpness-Aware Minimization (SAM) and mixup data augmentation. We achieve a new state-of-the-art on SoccerNet-v2, the largest soccer video dataset of its kind, with marked improvements in temporal localization. Additionally, our ablations show: the importance of predicting the temporal displacements; the trade-offs between the u-net and TE trunks; and the benefits of training with SAM and mixup.

We present a method for automated segmentation of the vasculature in retinal images. The method produces segmentations by classifying each image pixel as vessel or nonvessel, based on the pixel's feature vector. Feature vectors are composed of the pixel's intensity and two-dimensional Gabor wavelet transform responses taken at multiple scales. The Gabor wavelet is capable of tuning to specific frequencies, thus allowing noise filtering and vessel enhancement in a single step. We use a Bayesian classifier with class-conditional probability density functions (likelihoods) described as Gaussian mixtures, yielding a fast classification, while being able to model complex decision surfaces. The probability distributions are estimated based on a training set of labeled pixels obtained from manual segmentations. The method's performance is evaluated on publicly available DRIVE (Staal et al., 2004) and STARE (Hoover et al., 2000) databases of manually labeled images. On the DRIVE database, it achieves an area under the receiver operating characteristic curve of 0.9614, being slightly superior than that presented by state-of-the-art approaches. We are making our implementation available as open source MATLAB scripts for researchers interested in implementation details, evaluation, or development of methods.