Vision-and-language (VL) pre-training has proven to be highly effective on various VL downstream tasks. While recent work has shown that fully transformer-based VL models can be more efficient than previous region-feature-based methods, their performance on downstream tasks often degrades significantly. In this paper, we present METER, a Multimodal End-to-end TransformER framework, through which we investigate how to design and pre-train a fully transformer-based VL model in an end-to-end manner. Specifically, we dissect the model designs along multiple dimensions: vision encoders (e.g., CLIPViT, Swin transformer), text encoders (e.g., RoBERTa, DeBERTa), multimodal fusion module (e.g., merged attention vs. co-attention), architectural design (e.g., encoder-only vs. encoder-decoder), and pre-training objectives (e.g., masked image modeling). We conduct comprehensive experiments and provide insights on how to train a performant VL transformer while maintaining fast inference speed. Notably, our best model achieves an accuracy of 77.64% on the VQAv2 test-std set using only 4M images for pre-training, surpassing the state-of-the-art region-feature-based model by 1.04%, and outperforming the previous best fully transformer-based model by 1.6%. Code and models are released at https://github.com/zdou0830/METER.
In recent years, we have witnessed significant performance boost in the image captioning task based on vision-language pre-training (VLP). Scale is believed to be an important factor for this advance. However, most existing work only focuses on pre-training transformers with moderate sizes (e.g., 12 or 24 layers) on roughly 4 million images. In this paper, we present LEMON, a LargE-scale iMage captiONer, and provide the first empirical study on the scaling behavior of VLP for image captioning. We use the state-of-the-art VinVL model as our reference model, which consists of an image feature extractor and a transformer model, and scale the transformer both up and down, with model sizes ranging from 13 to 675 million parameters. In terms of data, we conduct experiments with up to 200 million image-text pairs which are automatically collected from web based on the alt attribute of the image (dubbed as ALT200M). Extensive analysis helps to characterize the performance trend as the model size and the pre-training data size increase. We also compare different training recipes, especially for training on large-scale noisy data. As a result, LEMON achieves new state of the arts on several major image captioning benchmarks, including COCO Caption, nocaps, and Conceptual Captions. We also show LEMON can generate captions with long-tail visual concepts when used in a zero-shot manner.
In this paper, we propose UNICORN, a vision-language (VL) model that unifies text generation and bounding box prediction into a single architecture. Specifically, we quantize each box into four discrete box tokens and serialize them as a sequence, which can be integrated with text tokens. We formulate all VL problems as a generation task, where the target sequence consists of the integrated text and box tokens. We then train a transformer encoder-decoder to predict the target in an auto-regressive manner. With such a unified framework and input-output format, UNICORN achieves comparable performance to task-specific state of the art on 7 VL benchmarks, covering the visual grounding, grounded captioning, visual question answering, and image captioning tasks. When trained with multi-task finetuning, UNICORN can approach different VL tasks with a single set of parameters, thus crossing downstream task boundary. We show that having a single model not only saves parameters, but also further boosts the model performance on certain tasks. Finally, UNICORN shows the capability of generalizing to new tasks such as ImageNet object localization.
Automated visual understanding of our diverse and open world demands computer vision models to generalize well with minimal customization for specific tasks, similar to human vision. Computer vision foundation models, which are trained on diverse, large-scale dataset and can be adapted to a wide range of downstream tasks, are critical for this mission to solve real-world computer vision applications. While existing vision foundation models such as CLIP, ALIGN, and Wu Dao 2.0 focus mainly on mapping images and textual representations to a cross-modal shared representation, we introduce a new computer vision foundation model, Florence, to expand the representations from coarse (scene) to fine (object), from static (images) to dynamic (videos), and from RGB to multiple modalities (caption, depth). By incorporating universal visual-language representations from Web-scale image-text data, our Florence model can be easily adapted for various computer vision tasks, such as classification, retrieval, object detection, VQA, image caption, video retrieval and action recognition. Moreover, Florence demonstrates outstanding performance in many types of transfer learning: fully sampled fine-tuning, linear probing, few-shot transfer and zero-shot transfer for novel images and objects. All of these properties are critical for our vision foundation model to serve general purpose vision tasks. Florence achieves new state-of-the-art results in majority of 44 representative benchmarks, e.g., ImageNet-1K zero-shot classification with top-1 accuracy of 83.74 and the top-5 accuracy of 97.18, 62.4 mAP on COCO fine tuning, 80.36 on VQA, and 87.8 on Kinetics-600.
In this paper, we propose a single UniFied transfOrmer (UFO), which is capable of processing either unimodal inputs (e.g., image or language) or multimodal inputs (e.g., the concatenation of the image and the question), for vision-language (VL) representation learning. Existing approaches typically design an individual network for each modality and/or a specific fusion network for multimodal tasks. To simplify the network architecture, we use a single transformer network and enforce multi-task learning during VL pre-training, which includes the image-text contrastive loss, image-text matching loss, and masked language modeling loss based on the bidirectional and the seq2seq attention mask. The same transformer network is used as the image encoder, the text encoder, or the fusion network in different pre-training tasks. Empirically, we observe less conflict among different tasks and achieve new state of the arts on visual question answering, COCO image captioning (cross-entropy optimization) and nocaps (in SPICE). On other downstream tasks, e.g., image-text retrieval, we also achieve competitive performance.
Motion deblurring has witnessed rapid development in recent years, and most of the recent methods address it by using deep learning techniques, with the help of different kinds of prior knowledge. Concerning that deblurring is essentially expected to improve the image sharpness, edge information can serve as an important prior. However, the edge has not yet been seriously taken into consideration in previous methods when designing deep models. To this end, we present a novel framework that incorporates edge prior knowledge into deep models, termed Edge Prior Augmented Networks (EPAN). EPAN has a content-based main branch and an edge-based auxiliary branch, which are constructed as a Content Deblurring Net (CDN) and an Edge Enhancement Net (EEN), respectively. EEN is designed to augment CDN in the deblurring process via an attentive fusion mechanism, where edge features are mapped as spatial masks to guide content features in a feature-based hierarchical manner. An edge-guided loss function is proposed to further regulate the optimization of EPAN by enforcing the focus on edge areas. Besides, we design a dual-camera-based image capturing setting to build a new dataset, Real Object Motion Blur (ROMB), with paired sharp and naturally blurry images of fast-moving cars, so as to better train motion deblurring models and benchmark the capability of motion deblurring algorithms in practice. Extensive experiments on the proposed ROMB and other existing datasets demonstrate that EPAN outperforms state-of-the-art approaches qualitatively and quantitatively.
Knowledge-based visual question answering (VQA) involves answering questions that require external knowledge not present in the image. Existing methods first retrieve knowledge from external resources, then reason over the selected knowledge, the input image, and question for answer prediction. However, this two-step approach could lead to mismatches that potentially limit the VQA performance. For example, the retrieved knowledge might be noisy and irrelevant to the question, and the re-embedded knowledge features during reasoning might deviate from their original meanings in the knowledge base (KB). To address this challenge, we propose PICa, a simple yet effective method that Prompts GPT3 via the use of Image Captions, for knowledge-based VQA. Inspired by GPT-3's power in knowledge retrieval and question answering, instead of using structured KBs as in previous work, we treat GPT-3 as an implicit and unstructured KB that can jointly acquire and process relevant knowledge. Specifically, we first convert the image into captions (or tags) that GPT-3 can understand, then adapt GPT-3 to solve the VQA task in a few-shot manner by just providing a few in-context VQA examples. We further boost performance by carefully investigating: (i) what text formats best describe the image content, and (ii) how in-context examples can be better selected and used. PICa unlocks the first use of GPT-3 for multimodal tasks. By using only 16 examples, PICa surpasses the supervised state of the art by an absolute +8.6 points on the OK-VQA dataset. We also benchmark PICa on VQAv2, where PICa also shows a decent few-shot performance.
This paper revisits human-object interaction (HOI) recognition at image level without using supervisions of object location and human pose. We name it detection-free HOI recognition, in contrast to the existing detection-supervised approaches which rely on object and keypoint detections to achieve state of the art. With our method, not only the detection supervision is evitable, but superior performance can be achieved by properly using image-text pre-training (such as CLIP) and the proposed Log-Sum-Exp Sign (LSE-Sign) loss function. Specifically, using text embeddings of class labels to initialize the linear classifier is essential for leveraging the CLIP pre-trained image encoder. In addition, LSE-Sign loss facilitates learning from multiple labels on an imbalanced dataset by normalizing gradients over all classes in a softmax format. Surprisingly, our detection-free solution achieves 60.5 mAP on the HICO dataset, outperforming the detection-supervised state of the art by 13.4 mAP
Imbalanced datasets widely exist in practice and area great challenge for training deep neural models with agood generalization on infrequent classes. In this work, wepropose a new rare-class sample generator (RSG) to solvethis problem. RSG aims to generate some new samplesfor rare classes during training, and it has in particularthe following advantages: (1) it is convenient to use andhighly versatile, because it can be easily integrated intoany kind of convolutional neural network, and it works wellwhen combined with different loss functions, and (2) it isonly used during the training phase, and therefore, no ad-ditional burden is imposed on deep neural networks duringthe testing phase. In extensive experimental evaluations, weverify the effectiveness of RSG. Furthermore, by leveragingRSG, we obtain competitive results on Imbalanced CIFARand new state-of-the-art results on Places-LT, ImageNet-LT, and iNaturalist 2018. The source code is available at https://github.com/Jianf-Wang/RSG.