Vision-language (VL) pre-training has recently received considerable attention. However, most existing end-to-end pre-training approaches either only aim to tackle VL tasks such as image-text retrieval, visual question answering (VQA) and image captioning that test high-level understanding of images, or only target region-level understanding for tasks such as phrase grounding and object detection. We present FIBER (Fusion-In-the-Backbone-based transformER), a new VL model architecture that can seamlessly handle both these types of tasks. Instead of having dedicated transformer layers for fusion after the uni-modal backbones, FIBER pushes multimodal fusion deep into the model by inserting cross-attention into the image and text backbones to better capture multimodal interactions. In addition, unlike previous work that is either only pre-trained on image-text data or on fine-grained data with box-level annotations, we present a two-stage pre-training strategy that uses both these kinds of data efficiently: (i) coarse-grained pre-training based on image-text data; followed by (ii) fine-grained pre-training based on image-text-box data. We conduct comprehensive experiments on a wide range of VL tasks, ranging from VQA, image captioning, and retrieval, to phrase grounding, referring expression comprehension, and object detection. Using deep multimodal fusion coupled with the two-stage pre-training, FIBER provides consistent performance improvements over strong baselines across all tasks, often outperforming methods using magnitudes more data. Code is released at https://github.com/microsoft/FIBER.
more »
« less
A Fast and Accurate One-Stage Approach to Visual Grounding
We propose a simple, fast, and accurate one-stage approach to visual grounding, inspired by the following insight. The performances of existing propose-and-rank twostage methods are capped by the quality of the region candidates they propose in the first stage — if none of the candidates could cover the ground truth region, there is no hope
in the second stage to rank the right region to the top. To
avoid this caveat, we propose a one-stage model that enables end-to-end joint optimization. The main idea is as
straightforward as fusing a text query’s embedding into the
YOLOv3 object detector, augmented by spatial features so
as to account for spatial mentions in the query. Despite being simple, this one-stage approach shows great potential
in terms of both accuracy and speed for both phrase localization and referring expression comprehension, according
to our experiments. Given these results along with careful
investigations into some popular region proposals, we advocate for visual grounding a paradigm shift from the conventional two-stage methods to the one-stage framework.
more »
« less
- PAR ID:
- 10169166
- Date Published:
- Journal Name:
- International Conference on Computer Vision
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
In most environments, the visual system is confronted with many relevant objects simultaneously. That is especially true during reading. However, behavioral data demonstrate that a serial bottleneck prevents recognition of more than one word at a time. We used fMRI to investigate how parallel spatial channels of visual processing converge into a serial bottleneck for word recognition. Participants viewed pairs of words presented simultaneously. We found that retinotopic cortex processed the two words in parallel spatial channels, one in each contralateral hemisphere. Responses were higher for attended than for ignored words but were not reduced when attention was divided. We then analyzed two word-selective regions along the occipitotemporal sulcus (OTS) of both hemispheres (subregions of the visual word form area, VWFA). Unlike retinotopic regions, each word-selective region responded to words on both sides of fixation. Nonetheless, a single region in the left hemisphere (posterior OTS) contained spatial channels for both hemifields that were independently modulated by selective attention. Thus, the left posterior VWFA supports parallel processing of multiple words. In contrast, activity in a more anterior word-selective region in the left hemisphere (mid OTS) was consistent with a single channel, showing (
i ) limited spatial selectivity, (ii ) no effect of spatial attention on mean response amplitudes, and (iii ) sensitivity to lexical properties of only one attended word. Therefore, the visual system can process two words in parallel up to a late stage in the ventral stream. The transition to a single channel is consistent with the observed bottleneck in behavior. -
null (Ed.)Understanding spatial relations (e.g., laptop on table) in visual input is important for both humans and robots. Existing datasets are insufficient as they lack large-scale, high-quality 3D ground truth information, which is critical for learning spatial relations. In this paper, we fill this gap by constructing Rel3D: the first large-scale, human-annotated dataset for grounding spatial relations in 3D. Rel3D enables quantifying the effectiveness of 3D information in predicting spatial relations on large-scale human data. Moreover, we propose minimally contrastive data collection---a novel crowdsourcing method for reducing dataset bias. The 3D scenes in our dataset come in minimally contrastive pairs: two scenes in a pair are almost identical, but a spatial relation holds in one and fails in the other. We empirically validate that minimally contrastive examples can diagnose issues with current relation detection models as well as lead to sample-efficient training. Code and data are available at https://github.com/princeton-vl/Rel3D.more » « less
-
Finding diagrams that contain a specific part or a similar part is important in many engineering tasks. In this search task, the query part is expected to match only a small region in a complex image.This paper investigates several local matching networks that explicitly model local region-to-region similarities. Deep convolutional neural networks extract local features and model local matching patterns. Spatial convolution is employed to cross-match local regions at different scale levels, addressing cases where the target part appears at a different scale, position, and/or angle. A gating network automatically learns region importance, removing noise from sparse areas and visual metadata in engineering diagrams. Experimental results show that local matching approaches are more effective for engineering diagram search than global matching approaches. Suppressing unimportant regions via the gating net-work enhances accuracy. Matching across different scales via spatial convolution substantially improves robustness to scale and rotation changes. A pipelined architecture efficiently searches a large collection of diagrams by using a simple local matching network to identify a small set of candidate images and a more sophisticated network with convolutional cross-scale matching to re-rank candidates.more » « less
-
Combining the preferences of many rankers into one single consensus ranking is critical for consequential applications from hiring and admissions to lending. While group fairness has been extensively studied for classification, group fairness in rankings and in particular rank aggregation remains in its infancy. Recent work introduced the concept of fair rank aggregation for combining rankings but restricted to the case when candidates have a single binary protected attribute, i.e., they fall into two groups only. Yet it remains an open problem how to create a consensus ranking that represents the preferences of all rankers while ensuring fair treatment for candidates with multiple protected attributes such as gender, race, and nationality. In this work, we are the first to define and solve this open Multi-attribute Fair Consensus Ranking (MFCR) problem. As a foundation, we design novel group fairness criteria for rankings, called MANI-Rank, ensuring fair treatment of groups defined by individual protected attributes and their intersection. Leveraging the MANI-Rank criteria, we develop a series of algorithms that for the first time tackle the MFCR problem. Our experimental study with a rich variety of consensus scenarios demonstrates our MFCR methodology is the only approach to achieve both intersectional and protected attribute fairness while also representing the preferences expressed through many base rankings. Our real-world case study on merit scholarships illustrates the effectiveness of our MFCR methods to mitigate bias across multiple protected attributes and their intersections.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICCV.2019.00478
