Inspired by the remarkable ability of the infant visual learning system, a recent study collected first-person images from children to analyze the `training data' that they receive. We conduct a follow-up study that investigates two additional directions. First, given that infants can quickly learn to recognize a new object without much supervision (i.e. few-shot learning), we limit the number of training images. Second, we investigate how children control the supervision signals they receive during learning based on hand manipulation of objects. Our experimental results suggest that supervision with hand manipulation is better than without hands, and the trend is consistent even when a small number of images is available.
Passive visual systems typically fail to recognize objects in the amodal setting where they are heavily occluded. In contrast, humans and other embodied agents have the ability to move in the environment, and actively control the viewing angle to better understand object shapes and semantics. In this work, we introduce the task of Embodied Visual Recognition (EVR): An agent is instantiated in a 3D environment close to an occluded target object, and is free to move in the environment to perform object classification, amodal object localization, and amodal object segmentation. To address this, we develop a new model called Embodied Mask R-CNN, for agents to learn to move strategically to improve their visual recognition abilities. We conduct experiments using the House3D environment. Experimental results show that: 1) agents with embodiment (movement) achieve better visual recognition performance than passive ones; 2) in order to improve visual recognition abilities, agents can learn strategical moving paths that are different from shortest paths.
Transferring knowledge across different datasets is an important approach to successfully train deep models with a small-scale target dataset or when few labeled instances are available. In this paper, we aim at developing a model that can generalize across multiple domain shifts, so that this model can adapt from a single source to multiple targets. This can be achieved by randomizing the generation of the data of various styles to mitigate the domain mismatch. First, we present a new adaptation to the CycleGAN model to produce stochastic style transfer between two image batches of different domains. Second, we enhance the classifier performance by using a self-ensembling technique with a teacher and student model to train on both original and generated data. Finally, we present experimental results on three datasets Office-31, Office-Home, and Visual Domain adaptation. The results suggest that selfensembling is better than simple data augmentation with the newly generated data and a single model trained this way can have the best performance across all different transfer tasks.
Neural sequence models are widely used to model time-series data. Equally ubiquitous is the usage of beam search (BS) as an approximate inference algorithm to decode output sequences from these models. BS explores the search space in a greedy left-right fashion retaining only the top-B candidates - resulting in sequences that differ only slightly from each other. Producing lists of nearly identical sequences is not only computationally wasteful but also typically fails to capture the inherent ambiguity of complex AI tasks. To overcome this problem, we propose Diverse Beam Search (DBS), an alternative to BS that decodes a list of diverse outputs by optimizing for a diversity-augmented objective. We observe that our method finds better top-1 solutions by controlling for the exploration and exploitation of the search space - implying that DBS is a better search algorithm. Moreover, these gains are achieved with minimal computational or memory over- head as compared to beam search. To demonstrate the broad applicability of our method, we present results on image captioning, machine translation and visual question generation using both standard quantitative metrics and qualitative human studies. Further, we study the role of diversity for image-grounded language generation tasks as the complexity of the image changes. We observe that our method consistently outperforms BS and previously proposed techniques for diverse decoding from neural sequence models.
CapsNet (Capsule Network) was first proposed by~\citet{capsule} and later another version of CapsNet was proposed by~\citet{emrouting}. CapsNet has been proved effective in modeling spatial features with much fewer parameters. However, the routing procedures in both papers are not well incorporated into the whole training process. The optimal number of routing procedure is misery which has to be found manually. To overcome this disadvantages of current routing procedures in CapsNet, we embed the routing procedure into the optimization procedure with all other parameters in neural networks, namely, make coupling coefficients in the routing procedure become completely trainable. We call it Generalized CapsNet (G-CapsNet). We implement both "full-connected" version of G-CapsNet and "convolutional" version of G-CapsNet. G-CapsNet achieves a similar performance in the dataset MNIST as in the original papers. We also test two capsule packing method (cross feature maps or with feature maps) from previous convolutional layers and see no evident difference. Besides, we also explored possibility of stacking multiple capsule layers. The code is shared on \hyperlink{https://github.com/chenzhenhua986/CAFFE-CapsNet}{CAFFE-CapsNet}.
Detecting small, densely distributed objects is a significant challenge: small objects often contain less distinctive information compared to larger ones, and finer-grained precision of bounding box boundaries are required. In this paper, we propose two techniques for addressing this problem. First, we estimate the likelihood that each pixel belongs to an object boundary rather than predicting coordinates of bounding boxes (as YOLO, Faster-RCNN and SSD do), by proposing a new architecture called Filter-Amplifier Networks (FANs). Second, we introduce a technique called Loss Boosting (LB) which attempts to soften the loss imbalance problem on each image. We test our algorithm on the problem of detecting electrical components on a new, realistic, diverse dataset of printed circuit boards (PCBs), as well as the problem of detecting vehicles in the Vehicle Detection in Aerial Imagery (VEDAI) dataset. Experiments show that our method works significantly better than current state-of-the-art algorithms with respect to accuracy, recall and average IoU.
A key problem in automatic analysis and understanding of scientific papers is to extract semantic information from non-textual paper components like figures, diagrams, tables, etc. Much of this work requires a very first preprocessing step: decomposing compound multi-part figures into individual subfigures. Previous work in compound figure separation has been based on manually designed features and separation rules, which often fail for less common figure types and layouts. Moreover, few implementations for compound figure decomposition are publicly available. This paper proposes a data driven approach to separate compound figures using modern deep Convolutional Neural Networks (CNNs) to train the separator in an end-to-end manner. CNNs eliminate the need for manually designing features and separation rules, but require a large amount of annotated training data. We overcome this challenge using transfer learning as well as automatically synthesizing training exemplars. We evaluate our technique on the ImageCLEF Medical dataset, achieving 85.9% accuracy and outperforming previous techniques. We have released our implementation as an easy-to-use Python library, aiming to promote further research in scientific figure mining.
Recent work in computer vision has yielded impressive results in automatically describing images with natural language. Most of these systems generate captions in a sin- gle language, requiring multiple language-specific models to build a multilingual captioning system. We propose a very simple technique to build a single unified model across languages, using artificial tokens to control the language, making the captioning system more compact. We evaluate our approach on generating English and Japanese captions, and show that a typical neural captioning architecture is capable of learning a single model that can switch between two different languages.
Given the progress in image recognition with recent data driven paradigms, it's still expensive to manually label a large training data to fit a convolutional neural network (CNN) model. This paper proposes a hybrid supervised-unsupervised method combining a pre-trained AlexNet with Latent Dirichlet Allocation (LDA) to extract image topics from both an unlabeled life-logging dataset and the COCO dataset. We generate the bag-of-words representations of an egocentric dataset from the softmax layer of AlexNet and use LDA to visualize the subject's living genre with duplicated images. We use a subset of COCO on 4 categories as ground truth, and define consistent rate to quantitatively analyze the performance of the method, it achieves 84% for consistent rate on average comparing to 18.75% from a raw CNN model. The method is capable of detecting false labels and multi-labels from COCO dataset. For scalability test, parallelization experiments are conducted with Harp-LDA on a Intel Knights Landing cluster: to extract 1,000 topic assignments for 241,035 COCO images, it takes 10 minutes with 60 threads.
Many practical perception systems exist within larger processes that include interactions with users or additional components capable of evaluating the quality of predicted solutions. In these contexts, it is beneficial to provide these oracle mechanisms with multiple highly likely hypotheses rather than a single prediction. In this work, we pose the task of producing multiple outputs as a learning problem over an ensemble of deep networks -- introducing a novel stochastic gradient descent based approach to minimize the loss with respect to an oracle. Our method is simple to implement, agnostic to both architecture and loss function, and parameter-free. Our approach achieves lower oracle error compared to existing methods on a wide range of tasks and deep architectures. We also show qualitatively that the diverse solutions produced often provide interpretable representations of task ambiguity.