This empirical study aims at improving the effectiveness of the standard 2-layer MLP projection head $g(\cdot)$ featured in the SimCLR framework through the use of pretrained autoencoder embeddings. Given a contrastive learning task with a largely unlabeled image classification dataset, we first train a shallow autoencoder architecture and extract its compressed representations contained in the encoder's embedding layer. After freezing the weights within this pretrained layer, we use it as a drop-in replacement for the input layer of SimCLR's default projector. Additionally, we also apply further architectural changes to the projector by decreasing its width and changing its activation function. The different projection heads are then used to contrastively train and evaluate a feature extractor $f(\cdot)$ following the SimCLR protocol, while also examining the performance impact of Z-score normalized datasets. Our experiments indicate that using a pretrained autoencoder embedding in the projector can not only increase classification accuracy by up to 2.9% or 1.7% on average but can also significantly decrease the dimensionality of the projection space. Our results also suggest, that using the sigmoid and tanh activation functions within the projector can outperform ReLU in terms of peak and average classification accuracy. When applying our presented projectors, then not applying Z-score normalization to datasets often increases peak performance. In contrast, the default projection head can benefit more from normalization. All experiments involving our pretrained projectors are conducted with frozen embeddings, since our test results indicate an advantage compared to using their non-frozen counterparts.
Despite extensive research, open-vocabulary segmentation methods still struggle to generalize across diverse domains. To reduce the computational cost of adapting Vision-Language Models (VLMs) while preserving their pre-trained knowledge, most methods freeze the VLMs for mask classification and train only the mask generator. However, our comprehensive analysis reveals a surprising insight: open-vocabulary segmentation is primarily bottlenecked by mask classification, not mask generation. This discovery prompts us to rethink the existing paradigm and explore an alternative approach. Instead of freezing the VLM, we propose to freeze the pre-trained mask generator and focus on optimizing the mask classifier. Building on the observation that VLMs pre-trained on global-pooled image-text features often fail to capture fine-grained semantics necessary for effective mask classification, we propose a novel Fine-grained Semantic Adaptation (FISA) method to address this limitation. FISA enhances the extracted visual features with fine-grained semantic awareness by explicitly integrating this crucial semantic information early in the visual encoding process. As our method strategically optimizes only a small portion of the VLM's parameters, it enjoys the efficiency of adapting to new data distributions while largely preserving the valuable VLM pre-trained knowledge. Extensive ablation studies confirm the superiority of our approach. Notably, FISA achieves new state-of-the-art results across multiple representative benchmarks, improving performance by up to +1.0 PQ and +3.0 mIoU and reduces training costs by nearly 5x compared to previous best methods. Our code and data will be made public.
We make distributed stochastic gradient descent faster by exchanging sparse updates instead of dense updates. Gradient updates are positively skewed as most updates are near zero, so we map the 99% smallest updates (by absolute value) to zero then exchange sparse matrices. This method can be combined with quantization to further improve the compression. We explore different configurations and apply them to neural machine translation and MNIST image classification tasks. Most configurations work on MNIST, whereas different configurations reduce convergence rate on the more complex translation task. Our experiments show that we can achieve up to 49% speed up on MNIST and 22% on NMT without damaging the final accuracy or BLEU.
Unmanned aircraft have decreased the cost required to collect remote sensing imagery, which has enabled researchers to collect high-spatial resolution data from multiple sensor modalities more frequently and easily. The increase in data will push the need for semantic segmentation frameworks that are able to classify non-RGB imagery, but this type of algorithmic development requires an increase in publicly available benchmark datasets with class labels. In this paper, we introduce a high-resolution multispectral dataset with image labels. This new benchmark dataset has been pre-split into training/testing folds in order to standardize evaluation and continue to push state-of-the-art classification frameworks for non-RGB imagery.
Actions as simple as grasping an object or navigating around it require a rich understanding of that object's 3D shape from a given viewpoint. In this paper we repurpose powerful learning machinery, originally developed for object classification, to discover image cues relevant for recovering the 3D shape of potentially unfamiliar objects. We cast the problem as one of local prediction of surface normals and global detection of 3D reflection symmetry planes, which open the door for extrapolating occluded surfaces from visible ones. We demonstrate that our method is able to recover accurate 3D shape information for classes of objects it was not trained on, in both synthetic and real images.
We address a challenging fine-grain classification problem: recognizing a font style from an image of text. In this task, it is very easy to generate lots of rendered font examples but very hard to obtain real-world labeled images. This real-to-synthetic domain gap caused poor generalization to new real data in previous methods (Chen et al. (2014)). In this paper, we refer to Convolutional Neural Networks, and use an adaptation technique based on a Stacked Convolutional Auto-Encoder that exploits unlabeled real-world images combined with synthetic data. The proposed method achieves an accuracy of higher than 80% (top-5) on a real-world dataset.
Deep neural networks have achieved a great success in solving many machine learning and computer vision problems. The main contribution of this paper is to develop a deep network based on Tucker tensor decomposition, and analyze its expressive power. It is shown that the expressiveness of Tucker network is more powerful than that of shallow network. In general, it is required to use an exponential number of nodes in a shallow network in order to represent a Tucker network. Experimental results are also given to compare the performance of the proposed Tucker network with hierarchical tensor network and shallow network, and demonstrate the usefulness of Tucker network in image classification problems.
Fully convolutional networks (FCNs) have been recently used for feature extraction and classification in image and speech recognition, where their inputs have been raw signal or other complicated features. Persian signature verification is done using conventional convolutional neural networks (CNNs). In this paper, we propose to use FCN for learning a robust feature extraction from the raw signature images. FCN can be considered as a variant of CNN where its fully connected layers are replaced with a global pooling layer. In the proposed manner, FCN inputs are raw signature images and convolution filter size is fixed. Recognition accuracy on UTSig database, shows that FCN with a global average pooling outperforms CNN.
We propose a novel method for training a neural network for image classification to reduce input data dynamically, in order to reduce the costs of training a neural network model. As Deep Learning tasks become more popular, their computational complexity increases, leading to more intricate algorithms and models which have longer runtimes and require more input data. The result is a greater cost on time, hardware, and environmental resources. By using data reduction techniques, we reduce the amount of work performed, and therefore the environmental impact of AI techniques, and with dynamic data reduction we show that accuracy may be maintained while reducing runtime by up to 50%, and reducing carbon emission proportionally.
We propose learnable polyphase sampling (LPS), a pair of learnable down/upsampling layers that enable truly shift-invariant and equivariant convolutional networks. LPS can be trained end-to-end from data and generalizes existing handcrafted downsampling layers. It is widely applicable as it can be integrated into any convolutional network by replacing down/upsampling layers. We evaluate LPS on image classification and semantic segmentation. Experiments show that LPS is on-par with or outperforms existing methods in both performance and shift consistency. For the first time, we achieve true shift-equivariance on semantic segmentation (PASCAL VOC), i.e., 100% shift consistency, outperforming baselines by an absolute 3.3%.
Greenhouse gases are pivotal drivers of climate change, necessitating precise quantification and source identification to foster mitigation strategies. We introduce GeoViT, a compact vision transformer model adept in processing satellite imagery for multimodal segmentation, classification, and regression tasks targeting CO2 and NO2 emissions. Leveraging GeoViT, we attain superior accuracy in estimating power generation rates, fuel type, plume coverage for CO2, and high-resolution NO2 concentration mapping, surpassing previous state-of-the-art models while significantly reducing model size. GeoViT demonstrates the efficacy of vision transformer architectures in harnessing satellite-derived data for enhanced GHG emission insights, proving instrumental in advancing climate change monitoring and emission regulation efforts globally.
Deep Neural Networks (DNN) are crucial in approximating nonlinear functions across diverse applications, ranging from image classification to control. Verifying specific input-output properties can be a highly challenging task due to the lack of a single, self-contained framework that allows a complete range of verification types. To this end, we present \texttt{ModelVerification.jl (MV)}, the first comprehensive, cutting-edge toolbox that contains a suite of state-of-the-art methods for verifying different types of DNNs and safety specifications. This versatile toolbox is designed to empower developers and machine learning practitioners with robust tools for verifying and ensuring the trustworthiness of their DNN models.
We have developed a novel activation function, named the Cauchy Activation Function. This function is derived from the Cauchy Integral Theorem in complex analysis and is specifically tailored for problems requiring high precision. This innovation has led to the creation of a new class of neural networks, which we call (Comple)XNet, or simply XNet. We will demonstrate that XNet is particularly effective for high-dimensional challenges such as image classification and solving Partial Differential Equations (PDEs). Our evaluations show that XNet significantly outperforms established benchmarks like MNIST and CIFAR-10 in computer vision, and offers substantial advantages over Physics-Informed Neural Networks (PINNs) in both low-dimensional and high-dimensional PDE scenarios.
This paper presents a comparison of the performance of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) in the task of multi-classifying images containing lesions of psoriasis and diseases similar to it. Models pre-trained on ImageNet were adapted to a specific data set. Both achieved high predictive metrics, but the ViTs stood out for their superior performance with smaller models. Dual Attention Vision Transformer-Base (DaViT-B) obtained the best results, with an f1-score of 96.4%, and is recommended as the most efficient architecture for automated psoriasis detection. This article reinforces the potential of ViTs for medical image classification tasks.
최소 중복 표현은 일반적으로 특징(feature) 공분산을 최소화하여 학습됩니다. 그러나 공분산 기반 방법은 선형적으로 상관관계가 없는 변수가 여전히 비선형 관계를 나타낼 수 있으므로 모든 의존성이나 중복성을 제거하지 못합니다. 이를 해결하기 위해 우리는 적대적 게임을 통해 특징 차원 간의 통계적 의존성을 최소화하는 미분 가능한 알고리즘인 ADM을 소개합니다. 여기서 보조 네트워크는 의존성을 식별하고 인코더는 이를 제거합니다. 우리는 전역 최적해에서 상호 독립성이 달성됨을 증명하고, 수렴(convergence)을 경험적으로 검증하며, 세 가지 잠재적 응용을 연구합니다: 주성분 분석(PCA)을 비선형 상관 제거로 확장, 이미지 분류(image classification)에서의 일반화 향상, 자기 지도 학습(self-supervised learning)에서의 차원 붕괴 방지. 통계적으로 독립적인 표현을 촉진함으로써, ADM은 다양한 응용 분야에 걸쳐 더 강인하고, 압축되며, 일반화 가능한 표현을 학습할 수 있는 기반을 마련합니다.
State-of-the-art object detectors are vulnerable to localized patch hiding attacks, where an adversary introduces a small adversarial patch to make detectors miss the detection of salient objects. The patch attacker can carry out a physical-world attack by printing and attaching an adversarial patch to the victim object. In this paper, we propose DetectorGuard as the first general framework for building provably robust object detectors against localized patch hiding attacks. DetectorGuard is inspired by recent advancements in robust image classification research; we ask: can we adapt robust image classifiers for robust object detection? Unfortunately, due to their task difference, an object detector naively adapted from a robust image classifier 1) may not necessarily be robust in the adversarial setting or 2) even maintain decent performance in the clean setting. To build a high-performance robust object detector, we propose an objectness explaining strategy: we adapt a robust image classifier to predict objectness for every image location and then explain each objectness using the bounding boxes predicted by a conventional object detector. If all objectness is well explained, we output the predictions made by the conventional object detector; otherwise, we issue an attack alert. Notably, 1) in the adversarial setting, we formally prove the end-to-end robustness of DetectorGuard on certified objects, i.e., it either detects the object or triggers an alert, against any patch hiding attacker within our threat model; 2) in the clean setting, we have almost the same performance as state-of-the-art object detectors. Our evaluation on the PASCAL VOC, MS COCO, and KITTI datasets further demonstrates that DetectorGuard achieves the first provable robustness against localized patch hiding attacks at a negligible cost (<1%) of clean performance.
Automatic image colourisation is the computer vision research path that studies how to colourise greyscale images (for restoration). Deep learning techniques improved image colourisation yielding astonishing results. These differ by various factors, such as structural differences, input types, user assistance, etc. Most of them, base the architectural structure on convolutional layers with no emphasis on layers specialised in object features extraction. We introduce a novel downsampling upsampling architecture named TUCaN (Tiny UCapsNet) that exploits the collaboration of convolutional layers and capsule layers to obtain a neat colourisation of entities present in every single image. This is obtained by enforcing collaboration among such layers by skip and residual connections. We pose the problem as a per pixel colour classification task that identifies colours as a bin in a quantized space. To train the network, in contrast with the standard end to end learning method, we propose the progressive learning scheme to extract the context of objects by only manipulating the learning process without changing the model. In this scheme, the upsampling starts from the reconstruction of low resolution images and progressively grows to high resolution images throughout the training phase. Experimental results on three benchmark datasets show that our approach with ImageNet10k dataset outperforms existing methods on standard quality metrics and achieves state of the art performances on image colourisation. We performed a user study to quantify the perceptual realism of the colourisation results demonstrating: that progressive learning let the TUCaN achieve better colours than the end to end scheme; and pointing out the limitations of the existing evaluation metrics.
Clouds in optical satellite images are a major concern since their presence hinders the ability to carry accurate analysis as well as processing. Presence of clouds also affects the image tasking schedule and results in wastage of valuable storage space on ground as well as space-based systems. Due to these reasons, deriving accurate cloud masks from optical remote-sensing images is an important task. Traditional methods such as threshold-based, spatial filtering for cloud detection in satellite images suffer from lack of accuracy. In recent years, deep learning algorithms have emerged as a promising approach to solve image segmentation problems as it allows pixel-level classification and semantic-level segmentation. In this paper, we introduce a deep-learning model based on hybrid transformer architecture for effective cloud mask generation named CLiSA - Cloud segmentation via Lipschitz Stable Attention network. In this context, we propose an concept of orthogonal self-attention combined with hierarchical cross attention model, and we validate its Lipschitz stability theoretically and empirically. We design the whole setup under adversarial setting in presence of Lov\'asz-Softmax loss. We demonstrate both qualitative and quantitative outcomes for multiple satellite image datasets including Landsat-8, Sentinel-2, and Cartosat-2s. Performing comparative study we show that our model performs preferably against other state-of-the-art methods and also provides better generalization in precise cloud extraction from satellite multi-spectral (MX) images. We also showcase different ablation studies to endorse our choices corresponding to different architectural elements and objective functions.
Data augmentation has become a pivotal tool in enhancing the performance of computer vision tasks, with the KeepOriginalAugment method emerging as a standout technique for its intelligent incorporation of salient regions within less prominent areas, enabling augmentation in both regions. Despite its success in image classification, its potential in addressing biases remains unexplored. In this study, we introduce an extension of the KeepOriginalAugment method, termed FaceKeepOriginalAugment, which explores various debiasing aspects-geographical, gender, and stereotypical biases-in computer vision models. By maintaining a delicate balance between data diversity and information preservation, our approach empowers models to exploit both diverse salient and non-salient regions, thereby fostering increased diversity and debiasing effects. We investigate multiple strategies for determining the placement of the salient region and swapping perspectives to decide which part undergoes augmentation. Leveraging the Image Similarity Score (ISS), we quantify dataset diversity across a range of datasets, including Flickr Faces HQ (FFHQ), WIKI, IMDB, Labelled Faces in the Wild (LFW), UTK Faces, and Diverse Dataset. We evaluate the effectiveness of FaceKeepOriginalAugment in mitigating gender bias across CEO, Engineer, Nurse, and School Teacher datasets, utilizing the Image-Image Association Score (IIAS) in convolutional neural networks (CNNs) and vision transformers (ViTs). Our findings shows the efficacy of FaceKeepOriginalAugment in promoting fairness and inclusivity within computer vision models, demonstrated by reduced gender bias and enhanced overall fairness. Additionally, we introduce a novel metric, Saliency-Based Diversity and Fairness Metric, which quantifies both diversity and fairness while handling data imbalance across various datasets.
There are great demands for automatically regulating inappropriate appearance of shocking firearm images in social media or identifying firearm types in forensics. Image retrieval techniques have great potential to solve these problems. To facilitate research in this area, we introduce Firearm 14k, a large dataset consisting of over 14,000 images in 167 categories. It can be used for both fine-grained recognition and retrieval of firearm images. Recent advances in image retrieval are mainly driven by fine-tuning state-of-the-art convolutional neural networks for retrieval task. The conventional single margin contrastive loss, known for its simplicity and good performance, has been widely used. We find that it performs poorly on the Firearm 14k dataset due to: (1) Loss contributed by positive and negative image pairs is unbalanced during training process. (2) A huge domain gap exists between this dataset and ImageNet. We propose to deal with the unbalanced loss by employing a double margin contrastive loss. We tackle the domain gap issue with a two-stage training strategy, where we first fine-tune the network for classification, and then fine-tune it for retrieval. Experimental results show that our approach outperforms the conventional single margin approach by a large margin (up to 88.5% relative improvement) and even surpasses the strong triplet-loss-based approach.