This paper presents OmniVL, a new foundation model to support both image-language and video-language tasks using one universal architecture. It adopts a unified transformer-based visual encoder for both image and video inputs, and thus can perform joint image-language and video-language pretraining. We demonstrate, for the first time, such a paradigm benefits both image and video tasks, as opposed to the conventional one-directional transfer (e.g., use image-language to help video-language). To this end, we propose a decoupled joint pretraining of image-language and video-language to effectively decompose the vision-language modeling into spatial and temporal dimensions and obtain performance boost on both image and video tasks. Moreover, we introduce a novel unified vision-language contrastive (UniVLC) loss to leverage image-text, video-text, image-label (e.g., image classification), video-label (e.g., video action recognition) data together, so that both supervised and noisily supervised pretraining data are utilized as much as possible. Without incurring extra task-specific adaptors, OmniVL can simultaneously support visual only tasks (e.g., image classification, video action recognition), cross-modal alignment tasks (e.g., image/video-text retrieval), and multi-modal understanding and generation tasks (e.g., image/video question answering, captioning). We evaluate OmniVL on a wide range of downstream tasks and achieve state-of-the-art or competitive results with similar model size and data scale.
This paper focuses on the task of survival time analysis for lung cancer. Although much progress has been made in this problem in recent years, the performance of existing methods is still far from satisfactory. Traditional and some deep learning-based survival time analyses for lung cancer are mostly based on textual clinical information such as staging, age, histology, etc. Unlike existing methods that predicting on the single modality, we observe that a human clinician usually takes multimodal data such as text clinical data and visual scans to estimate survival time. Motivated by this, in this work, we contribute a smart cross-modality network for survival analysis network named Lite-ProSENet that simulates a human's manner of decision making. Extensive experiments were conducted using data from 422 NSCLC patients from The Cancer Imaging Archive (TCIA). The results show that our Lite-ProSENet outperforms favorably again all comparison methods and achieves the new state of the art with the 89.3% on concordance. The code will be made publicly available.
In this paper, we study the problem of identifying logos of business brands in natural scenes in an open-set one-shot setting. This problem setup is significantly more challenging than traditionally-studied 'closed-set' and 'large-scale training samples per category' logo recognition settings. We propose a novel multi-view textual-visual encoding framework that encodes text appearing in the logos as well as the graphical design of the logos to learn robust contrastive representations. These representations are jointly learned for multiple views of logos over a batch and thereby they generalize well to unseen logos. We evaluate our proposed framework for cropped logo verification, cropped logo identification, and end-to-end logo identification in natural scene tasks; and compare it against state-of-the-art methods. Further, the literature lacks a 'very-large-scale' collection of reference logo images that can facilitate the study of one-hundred thousand-scale logo identification. To fill this gap in the literature, we introduce Wikidata Reference Logo Dataset (WiRLD), containing logos for 100K business brands harvested from Wikidata. Our proposed framework that achieves an area under the ROC curve of 91.3% on the QMUL-OpenLogo dataset for the verification task, outperforms state-of-the-art methods by 9.1% and 2.6% on the one-shot logo identification task on the Toplogos-10 and the FlickrLogos32 datasets, respectively. Further, we show that our method is more stable compared to other baselines even when the number of candidate logos is on a 100K scale.
Podcasts have emerged as a massively consumed online content, notably due to wider accessibility of production means and scaled distribution through large streaming platforms. Categorization systems and information access technologies typically use topics as the primary way to organize or navigate podcast collections. However, annotating podcasts with topics is still quite problematic because the assigned editorial genres are broad, heterogeneous or misleading, or because of data challenges (e.g. short metadata text, noisy transcripts). Here, we assess the feasibility to discover relevant topics from podcast metadata, titles and descriptions, using topic modeling techniques for short text. We also propose a new strategy to leverage named entities (NEs), often present in podcast metadata, in a Non-negative Matrix Factorization (NMF) topic modeling framework. Our experiments on two existing datasets from Spotify and iTunes and Deezer, a new dataset from an online service providing a catalog of podcasts, show that our proposed document representation, NEiCE, leads to improved topic coherence over the baselines. We release the code for experimental reproducibility of the results.
Coming up with effective ad text is a time consuming process, and particularly challenging for small businesses with limited advertising experience. When an inexperienced advertiser onboards with a poorly written ad text, the ad platform has the opportunity to detect low performing ad text, and provide improvement suggestions. To realize this opportunity, we propose an ad text strength indicator (TSI) which: (i) predicts the click-through-rate (CTR) for an input ad text, (ii) fetches similar existing ads to create a neighborhood around the input ad, (iii) and compares the predicted CTRs in the neighborhood to declare whether the input ad is strong or weak. In addition, as suggestions for ad text improvement, TSI shows anonymized versions of superior ads (higher predicted CTR) in the neighborhood. For (i), we propose a BERT based text-to-CTR model trained on impressions and clicks associated with an ad text. For (ii), we propose a sentence-BERT based semantic-ad-similarity model trained using weak labels from ad campaign setup data. Offline experiments demonstrate that our BERT based text-to-CTR model achieves a significant lift in CTR prediction AUC for cold start (new) advertisers compared to bag-of-words based baselines. In addition, our semantic-textual-similarity model for similar ads retrieval achieves a precision@1 of 0.93 (for retrieving ads from the same product category); this is significantly higher compared to unsupervised TF-IDF, word2vec, and sentence-BERT baselines. Finally, we share promising online results from advertisers in the Yahoo (Verizon Media) ad platform where a variant of TSI was implemented with sub-second end-to-end latency.
This paper proposes a multimodal emotion recognition system, VIsual Spoken Textual Additive Net (VISTA Net), to classify the emotions reflected by a multimodal input containing image, speech, and text into discrete classes. A new interpretability technique, K-Average Additive exPlanation (KAAP), has also been developed to identify the important visual, spoken, and textual features leading to predicting a particular emotion class. The VISTA Net fuses the information from image, speech & text modalities using a hybrid of early and late fusion. It automatically adjusts the weights of their intermediate outputs while computing the weighted average without human intervention. The KAAP technique computes the contribution of each modality and corresponding features toward predicting a particular emotion class. To mitigate the insufficiency of multimodal emotion datasets labeled with discrete emotion classes, we have constructed a large-scale IIT-R MMEmoRec dataset consisting of real-life images, corresponding speech & text, and emotion labels ('angry,' 'happy,' 'hate,' and 'sad.'). The VISTA Net has resulted in 95.99% emotion recognition accuracy on considering image, speech, and text modalities, which is better than the performance on considering the inputs of any one or two modalities.
The zero-shot scenario for speech generation aims at synthesizing a novel unseen voice with only one utterance of the target speaker. Although the challenges of adapting new voices in zero-shot scenario exist in both stages -- acoustic modeling and vocoder, previous works usually consider the problem from only one stage. In this paper, we extend our previous Glow-WaveGAN to Glow-WaveGAN 2, aiming to solve the problem from both stages for high-quality zero-shot text-to-speech and any-to-any voice conversion. We first build a universal WaveGAN model for extracting latent distribution $p(z)$ of speech and reconstructing waveform from it. Then a flow-based acoustic model only needs to learn the same $p(z)$ from texts, which naturally avoids the mismatch between the acoustic model and the vocoder, resulting in high-quality generated speech without model fine-tuning. Based on a continuous speaker space and the reversible property of flows, the conditional distribution can be obtained for any speaker, and thus we can further conduct high-quality zero-shot speech generation for new speakers. We particularly investigate two methods to construct the speaker space, namely pre-trained speaker encoder and jointly-trained speaker encoder. The superiority of Glow-WaveGAN 2 has been proved through TTS and VC experiments conducted on LibriTTS corpus and VTCK corpus.
This paper presents Z-Code++, a new pre-trained language model optimized for abstractive text summarization. The model extends the state of the art encoder-decoder model using three techniques. First, we use a two-phase pre-training process to improve model's performance on low-resource summarization tasks. The model is first pre-trained using text corpora for language understanding, and then is continually pre-trained on summarization corpora for grounded text generation. Second, we replace self-attention layers in the encoder with disentangled attention layers, where each word is represented using two vectors that encode its content and position, respectively. Third, we use fusion-in-encoder, a simple yet effective method of encoding long sequences in a hierarchical manner. Z-Code++ creates new state of the art on 9 out of 13 text summarization tasks across 5 languages. Our model is parameter-efficient in that it outperforms the 600x larger PaLM-540B on XSum, and the finetuned 200x larger GPT3-175B on SAMSum. In zero-shot and few-shot settings, our model substantially outperforms the competing models.
Recently, large-scale pre-trained Vision-and-Language (VL) foundation models have demonstrated remarkable capabilities in many zero-shot downstream tasks, achieving competitive results for recognizing objects defined by as little as short text prompts. However, it has also been shown that VL models are still brittle in Structured VL Concept (SVLC) reasoning, such as the ability to recognize object attributes, states, and inter-object relations. This leads to reasoning mistakes, which need to be corrected as they occur by teaching VL models the missing SVLC skills; often this must be done using private data where the issue was found, which naturally leads to a data-free continual (no task-id) VL learning setting. In this work, we introduce the first Continual Data-Free Structured VL Concepts Learning (ConStruct-VL) benchmark and show it is challenging for many existing data-free CL strategies. We, therefore, propose a data-free method comprised of a new approach of Adversarial Pseudo-Replay (APR) which generates adversarial reminders of past tasks from past task models. To use this method efficiently, we also propose a continual parameter-efficient Layered-LoRA (LaLo) neural architecture allowing no-memory-cost access to all past models at train time. We show this approach outperforms all data-free methods by as much as ~7% while even matching some levels of experience-replay (prohibitive for applications where data-privacy must be preserved).
Vision-language (V+L) pretraining models have achieved great success in supporting multimedia applications by understanding the alignments between images and text. While existing vision-language pretraining models primarily focus on understanding objects in images or entities in text, they often ignore the alignment at the level of events and their argument structures. % In this work, we propose a contrastive learning framework to enforce vision-language pretraining models to comprehend events and associated argument (participant) roles. To achieve this, we take advantage of text information extraction technologies to obtain event structural knowledge, and utilize multiple prompt functions to contrast difficult negative descriptions by manipulating event structures. We also design an event graph alignment loss based on optimal transport to capture event argument structures. In addition, we collect a large event-rich dataset (106,875 images) for pretraining, which provides a more challenging image retrieval benchmark to assess the understanding of complicated lengthy sentences. Experiments show that our zero-shot CLIP-Event outperforms the state-of-the-art supervised model in argument extraction on Multimedia Event Extraction, achieving more than 5\% absolute F-score gain in event extraction, as well as significant improvements on a variety of downstream tasks under zero-shot settings.