Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. 
The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

The high-speed feature detection is still very demanding for many computer vision applications. In this paper, the Harris and KLT corner detectors are redesigned to reduce the complexity of the algorithm. The complexity of Harris and KLT corner detectors are reduced by using the box kernel, the integral image and efficient non-maximum suppression, achieving complexity reduction by a factor of 8. For the image of size 1000×700, our method costs only 74ms on a commodity CPU with 2GHz and 1GB RAM.

Lococo: low complexity corner detector

Deformable Parts Models (DPM) are the current state-of-the-art for object detection. Nevertheless they seem sub-optimal in the representation of deformations. Object deformations are often continuous and not confined to big parts. Therefore we propose to replace the DPM star model based on big parts by a deformation field. This consists of a grid of small parts connected with pairwise constraints which can better handle continuous deformations. The naive application of this model for object detection would consist of a bounded sliding window approach: for each possible location of the image the best part configuration within a limited bound around this location is found. This is computationally very expensive.Instead, we propose a different inference procedure, where an iterative image-level search finds the best object hypothesis. We show that this approach is faster than bounded sliding windows yet produces comparable accuracy. Experiments further show that the deformation field can better approximate real object deformations and therefore, for certain classes, produces even better detection accuracy than state-of-the-art DPM. Finally, the same approach is adapted to model-free tracking, showing improved accuracy also in this case.

/pdf/an-elastic-deformation-field-model-for-object-detection-and-26pddmc1rb.pdf

An Elastic Deformation Field Model for Object Detection and Tracking

This paper introduces speech-based visual question answering (VQA), the task of generating an answer given an image and a spoken question. Two methods are studied: an end-to-end, deep neural network that directly uses audio waveforms as input versus a pipelined approach that performs ASR (Automatic Speech Recognition) on the question, followed by text-based visual question answering. Furthermore, we investigate the robustness of both methods by injecting various levels of noise into the spoken question and find both methods to be tolerate noise at similar levels.

Speech-Based Visual Question Answering.

So far, most image mining was based on interactive querying. Although such querying will remain important in the future, several applications need image mining at such wide scales that it has to run automatically. This adds an additional level to the problem, namely to apply appropriate further processing to different types of images, and to decide on such processing automatically as well. This paper touches on those issues in that we discuss the processing of landmark images and of images coming from webcams. The first part deals with the automated collection of images of landmarks, which are then also automatically annotated and enriched with Wikipedia information. The target application is that users photograph landmarks with their mobile phones or PDAs, and automatically get information about them. Similarly, users can get images in their photo albums annotated automatically. The object of interest can also be automatically delineated in the images. The pipeline we propose actually retrieves more images than manual keyword input would produce. The second part of the paper deals with an entirely different source of image data, but one that also produces massive amounts (although typically not archived): webcams. They produce images at a single location, but rather continuously and over extended periods of time. We propose an approach to summarize data coming from webcams. This data handling is quite different from that applied to the landmark images.

Mining from large image sets

The task of unsupervised semantic segmentation aims to cluster pixels into semantically meaningful groups. Specifically, pixels assigned to the same cluster should share high-level semantic properties like their object or part category. This paper presents MaskDistill: a novel framework for unsupervised semantic segmentation based on three key ideas. First, we advocate a data-driven strategy to generate object masks that serve as a pixel grouping prior for semantic segmentation. This approach omits handcrafted priors, which are often designed for specific scene compositions and limit the applicability of competing frameworks. Second, MaskDistill clusters the object masks to obtain pseudo-ground-truth for training an initial object segmentation model. Third, we leverage this model to filter out low-quality object masks. This strategy mitigates the noise in our pixel grouping prior and results in a clean collection of masks which we use to train a final segmentation model. By combining these components, we can considerably outperform previous works for unsupervised semantic segmentation on PASCAL (+11% mIoU) and COCO (+4% mask AP50). Interestingly, as opposed to existing approaches, our framework does not latch onto low-level image cues and is not limited to object-centric datasets. The code and models will be made available.

Luc Van Gool

Papers

Lococo: low complexity corner detector

An Elastic Deformation Field Model for Object Detection and Tracking

Speech-Based Visual Question Answering.

Mining from large image sets

Discovering Object Masks with Transformers for Unsupervised Semantic Segmentation