Assignació d’imatges i espais conceptuals: recuperació d’imatges basada en croquis
- Autores: Sounak Dey
- Directores de la Tesis: Josep Llados Canet (dir. tes.), Umapada Pal (codir. tes.)
- Lectura: En la Universitat Autònoma de Barcelona ( España ) en 2020
- Idioma: español
- Tribunal Calificador de la Tesis: Francesc Moreno Noguer (presid.), Oriol Ramos Terrades (secret.), Eric Anquetil (voc.)
- Programa de doctorado: Programa de Doctorado en Informática por la Universidad Autónoma de Barcelona
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
The deluge of visual content on the Internet – from user-generated content to commercial image collections - motivates intuitive new methods for searching digital image content: how can we find certain images in a database of millions? Sketch-based image retrieval (SBIR) is an emerging research topic in which a free-hand drawing can be used to visually query photographic images. SBIR is aligned to emerging trends for visual content consumption on mobile touch-screen based devices, for which gestural interactions such as sketch are a natural alternative to textual input.
This thesis presents several contributions to the literature of SBIR. First, we propose a cross-modal learning framework that maps both sketches and text into a joint embedding space invariant to depictive style, while preserving semantics. The resulting embedding enables direct comparison and search between sketches/text and images and is based upon a multi-branch convolutional neural network (CNN) trained using unique training schemes. The deeply learned embedding is shown to yield state-of-art retrieval performance on several SBIR benchmarks.
Second, we propose an approach for multi-modal image retrieval in multi-labelled images. A multi-modal deep network architecture is formulated to jointly model sket-ches and text as input query modalities into a common embedding space, which is then further aligned with the image feature space.
Our architecture also relies on a salient object detection through a supervised LSTM-based visual attention model lear-ned from convolutional features. Both the alignment between the queries and the image and the supervision of the attention on the images are obtained by generalizing the Hungarian Algorithm using different loss functions. This permits encoding the object-based features and its alignment with the query irrespective of the availability of the co-occurrence of different objects in the training set. We validate the performance of our approach on standard single/multi-object datasets, showing state-of-the art performance in every SBIR dataset.
Third, we investigate the problem of zero-shot sketch-based image retrieval (ZS-SBIR), where human sketches are used as queries to conduct retrieval of photos from unseen categories. We importantly advance prior arts by proposing a novel ZS-SBIR scenario that represents a firm step forward in its practical application. The new setting uniquely recognizes two important yet often neglected challenges of practical ZS-SBIR, (i) the large domain gap between amateur sketch and photo, and (ii) the necessity for moving towards large-scale retrieval. We first contribute to the community a novel ZS-SBIR dataset, QuickDraw-Extended, that consists of $330,000$ sketches and $204,000$ photos spanning across 110 categories. Highly abstract amateur human sketches are purposefully sourced to maximize the domain gap, instead of ones included in existing datasets that can often be semi-photorealistic. We then formulate a ZS-SBIR framework to jointly model sketches and photos into a common embedding space. A novel strategy to mine the mutual information among domains is specifically engineered to alleviate the domain gap. External semantic knowledge is further embedded to aid semantic transfer. We show that, rather surprisingly, retrieval performance significantly outperforms that of state-of-the-art on existing datasets that can already be achieved using a reduced version of our model. We further demonstrate the superior performance of our full model by comparing with a number of alternatives on the newly proposed dataset.
