Università degli studi di Modena e Reggio Emilia

Since their introduction the Trasformer architectures emerged as the dominating architectures for both natural language processing and, more recently, computer vision applications. An intrinsic limitation of this family of “fully-attentive” architectures arises from the computation of the dot-product attention, which grows both in memory consumption and number of operations as O(n^2) where n stands for the input sequence length, thus limiting the applications that require modeling very long sequences. Several approaches have been proposed so far in the literature to mitigate this issue, with varying degrees of success. Our idea takes inspiration from the world of lossy data compression (such as the JPEG algorithm) to derive an approximation of the attention module by leveraging the properties of the Discrete Cosine Transform. An extensive section of experiments shows that our method takes up less memory for the same performance, while also drastically reducing inference time. Moreover, we assume that the results of our research might serve as a starting point for a broader family of deep neural models with reduced memory footprint. The implementation will be made publicly available at https://github.com/cscribano/DCT-Former-Public.

With the recent advancements in the field of diffusion generative models, it has been shown that defining the generative process in the latent space of a powerful pretrained autoencoder can offer substantial advantages. This approach, by abstracting away imperceptible image details and introducing substantial spatial compression, renders the learning of the generative process more manageable while significantly reducing computational and memory demands. In this work, we propose to replace autoencoder coding with a model-based coding scheme based on traditional lossy image compression techniques; this choice not only further diminishes computational expenses but also allows us to probe the boundaries of latent-space image generation. Our objectives culminate in the proposal of a valuable approximation for training continuous diffusion models within a discrete space, accompanied by enhancements to the generative model for categorical values. Beyond the good results obtained for the problem at hand, we believe that the proposed work holds promise for enhancing the adaptability of generative diffusion models across diverse data types beyond the realm of imagery.

Combining Natural Language with Vision represents a unique and interesting challenge in the domain of Artificial Intelligence. The AI City Challenge Track 5 for Natural Language-Based Vehicle Retrieval focuses on the problem of combining visual and textual information, applied to a smart-city use case. In this paper, we present All You Can Embed (AYCE), a modular solution to correlate single-vehicle tracking sequences with natural language. The main building blocks of the proposed architecture are (i) BERT to provide an embedding of the textual descriptions, (ii) a convolutional backbone along with a Transformer model to embed the visual information. For the training of the retrieval model, a variation of the Triplet Margin Loss is proposed to learn a distance measure between the visual and language embeddings. The code is publicly available at https://github.com/cscribano/AYCE_2021.