Vision-language models (VLMs) that are pre-trained on large-scale image-text pairs have demonstrated impressive transferability on a wide range of visual tasks. Transferring knowledge from such powerful pre-trained VLMs is emerging as a promising direction for building effective video recognition models. However, the current exploration is still limited. In our opinion, the greatest charm of pre-trained vision-language models is to build a bridge between visual and textual domains. In this paper, we present a novel framework called BIKE which utilizes the cross-modal bridge to explore bidirectional knowledge: i) We propose a Video Attribute Association mechanism which leverages the Video-to-Text knowledge to generate textual auxiliary attributes to complement video recognition. ii) We also present a Temporal Concept Spotting mechanism which uses the Text-to-Video expertise to capture temporal saliency in a parameter-free manner to yield enhanced video representation. The extensive studies on popular video datasets (ie, Kinetics-400 & 600, UCF-101, HMDB-51 and ActivityNet) show that our method achieves state-of-the-art performance in most recognition scenarios, eg, general, zero-shot, and few-shot video recognition. To the best of our knowledge, our best model achieves a state-of-the-art accuracy of 88.4% on challenging Kinetics-400 with the released CLIP pre-trained model.
We demonstrate a physics-aware transformer for feature-based data fusion from cameras with diverse resolution, color spaces, focal planes, focal lengths, and exposure. We also demonstrate a scalable solution for synthetic training data generation for the transformer using open-source computer graphics software. We demonstrate image synthesis on arrays with diverse spectral responses, instantaneous field of view and frame rate.
Learning with noisy label (LNL) is a classic problem that has been extensively studied for image tasks, but much less for video in the literature. A straightforward migration from images to videos without considering the properties of videos, such as computational cost and redundant information, is not a sound choice. In this paper, we propose two new strategies for video analysis with noisy labels: 1) A lightweight channel selection method dubbed as Channel Truncation for feature-based label noise detection. This method selects the most discriminative channels to split clean and noisy instances in each category; 2) A novel contrastive strategy dubbed as Noise Contrastive Learning, which constructs the relationship between clean and noisy instances to regularize model training. Experiments on three well-known benchmark datasets for video classification show that our proposed tru{\bf N}cat{\bf E}-split-contr{\bf A}s{\bf T} (NEAT) significantly outperforms the existing baselines. By reducing the dimension to 10\% of it, our method achieves over 0.4 noise detection F1-score and 5\% classification accuracy improvement on Mini-Kinetics dataset under severe noise (symmetric-80\%). Thanks to Noise Contrastive Learning, the average classification accuracy improvement on Mini-Kinetics and Sth-Sth-V1 is over 1.6\%.
In this paper, we propose an end-to-end SpA-Former to recover a shadow-free image from a single shaded image. Unlike traditional methods that require two steps for shadow detection and then shadow removal, the SpA-Former unifies these steps into one, which is a one-stage network capable of directly learning the mapping function between shadows and no shadows, it does not require a separate shadow detection. Thus, SpA-former is adaptable to real image de-shadowing for shadows projected on different semantic regions. SpA-Former consists of transformer layer and a series of joint Fourier transform residual blocks and two-wheel joint spatial attention. The network in this paper is able to handle the task while achieving a very fast processing efficiency. Our code is relased on https://github.com/ zhangbaijin/Spatial-Transformer-shadow-removal
Object detection is a computer vision task of predicting a set of bounding boxes and category labels for each object of interest in a given image. The category is related to a linguistic symbol such as 'dog' or 'person' and there should be relationships among them. However the object detector only learns to classify the categories and does not treat them as the linguistic symbols. Multi-modal models often use the pre-trained object detector to extract object features from the image, but the models are separated from the detector and the extracted visual features does not change with their linguistic input. We rethink the object detection as a vision-and-language reasoning task. We then propose targeted detection task, where detection targets are given by a natural language and the goal of the task is to detect only all the target objects in a given image. There are no detection if the target is not given. Commonly used modern object detectors have many hand-designed components like anchor and it is difficult to fuse the textual inputs into the complex pipeline. We thus propose Language-Targeted Detector (LTD) for the targeted detection based on a recently proposed Transformer-based detector. LTD is a encoder-decoder architecture and our conditional decoder allows the model to reason about the encoded image with the textual input as the linguistic context. We evaluate detection performances of LTD on COCO object detection dataset and also show that our model improves the detection results with the textual input grounding to the visual object.
Multimodal relation extraction is an essential task for knowledge graph construction. In this paper, we take an in-depth empirical analysis that indicates the inaccurate information in the visual scene graph leads to poor modal alignment weights, further degrading performance. Moreover, the visual shuffle experiments illustrate that the current approaches may not take full advantage of visual information. Based on the above observation, we further propose a strong baseline with an implicit fine-grained multimodal alignment based on Transformer for multimodal relation extraction. Experimental results demonstrate the better performance of our method. Codes are available at https://github.com/zjunlp/DeepKE/tree/main/example/re/multimodal.
Deep learning-based single image super-resolution (SISR) approaches have drawn much attention and achieved remarkable success on modern advanced GPUs. However, most state-of-the-art methods require a huge number of parameters, memories, and computational resources, which usually show inferior inference times when applying them to current mobile device CPUs/NPUs. In this paper, we propose a simple plain convolution network with a fast nearest convolution module (NCNet), which is NPU-friendly and can perform a reliable super-resolution in real-time. The proposed nearest convolution has the same performance as the nearest upsampling but is much faster and more suitable for Android NNAPI. Our model can be easily deployed on mobile devices with 8-bit quantization and is fully compatible with all major mobile AI accelerators. Moreover, we conduct comprehensive experiments on different tensor operations on a mobile device to illustrate the efficiency of our network architecture. Our NCNet is trained and validated on the DIV2K 3x dataset, and the comparison with other efficient SR methods demonstrated that the NCNet can achieve high fidelity SR results while using fewer inference times. Our codes and pretrained models are publicly available at \url{https://github.com/Algolzw/NCNet}.
Text erasure from an image is helpful for various tasks such as image editing and privacy preservation. In this paper, we present TPFNet, a novel one-stage (end-toend) network for text removal from images. Our network has two parts: feature synthesis and image generation. Since noise can be more effectively removed from low-resolution images, part 1 operates on low-resolution images. The output of part 1 is a low-resolution text-free image. Part 2 uses the features learned in part 1 to predict a high-resolution text-free image. In part 1, we use "pyramidal vision transformer" (PVT) as the encoder. Further, we use a novel multi-headed decoder that generates a high-pass filtered image and a segmentation map, in addition to a text-free image. The segmentation branch helps locate the text precisely, and the high-pass branch helps in learning the image structure. To precisely locate the text, TPFNet employs an adversarial loss that is conditional on the segmentation map rather than the input image. On Oxford, SCUT, and SCUT-EnsText datasets, our network outperforms recently proposed networks on nearly all the metrics. For example, on SCUT-EnsText dataset, TPFNet has a PSNR (higher is better) of 39.0 and text-detection precision (lower is better) of 21.1, compared to the best previous technique, which has a PSNR of 32.3 and precision of 53.2. The source code can be obtained from https://github.com/CandleLabAI/TPFNet
We study the problem of semantic segmentation calibration. For image classification, lots of existing solutions are proposed to alleviate model miscalibration of confidence. However, to date, confidence calibration research on semantic segmentation is still limited. We provide a systematic study on the calibration of semantic segmentation models and propose a simple yet effective approach. First, we find that model capacity, crop size, multi-scale testing, and prediction correctness have impact on calibration. Among them, prediction correctness, especially misprediction, is more important to miscalibration due to over-confidence. Next, we propose a simple, unifying, and effective approach, namely selective scaling, by separating correct/incorrect prediction for scaling and more focusing on misprediction logit smoothing. Then, we study popular existing calibration methods and compare them with selective scaling on semantic segmentation calibration. We conduct extensive experiments with a variety of benchmarks on both in-domain and domain-shift calibration, and show that selective scaling consistently outperforms other methods.
Neural Radiance Fields (NeRFs) are emerging as a ubiquitous scene representation that allows for novel view synthesis. Increasingly, NeRFs will be shareable with other people. Before sharing a NeRF, though, it might be desirable to remove personal information or unsightly objects. Such removal is not easily achieved with the current NeRF editing frameworks. We propose a framework to remove objects from a NeRF representation created from an RGB-D sequence. Our NeRF inpainting method leverages recent work in 2D image inpainting and is guided by a user-provided mask. Our algorithm is underpinned by a confidence based view selection procedure. It chooses which of the individual 2D inpainted images to use in the creation of the NeRF, so that the resulting inpainted NeRF is 3D consistent. We show that our method for NeRF editing is effective for synthesizing plausible inpaintings in a multi-view coherent manner. We validate our approach using a new and still-challenging dataset for the task of NeRF inpainting.