We present ALF (Advertiser Large Foundation model), a multi-modal transformer architecture for understanding advertiser behavior and intent across text, image, video, and structured data modalities. Through contrastive learning and multi-task optimization, ALF creates unified advertiser representations that capture both content and behavioral patterns. Our model achieves state-of-the-art performance on critical tasks including fraud detection, policy violation identification, and advertiser similarity matching. In production deployment, ALF demonstrates significant real-world impact by delivering simultaneous gains in both precision and recall, for instance boosting recall by over 40 percentage points on one critical policy and increasing precision to 99.8% on another. The architecture's effectiveness stems from its novel combination of multi-modal transformations, inter-sample attention mechanism, spectrally normalized projections, and calibrated probabilistic outputs.

The goal of No-Reference Image Quality Assessment (NR-IQA) is to estimate the perceptual image quality in accordance with subjective evaluations, it is a complex and unsolved problem due to the absence of the pristine reference image. In this paper, we propose a novel model to address the NR-IQA task by leveraging a hybrid approach that benefits from Convolutional Neural Networks (CNNs) and self-attention mechanism in Transformers to extract both local and non-local features from the input image. We capture local structure information of the image via CNNs, then to circumvent the locality bias among the extracted CNNs features and obtain a non-local representation of the image, we utilize Transformers on the extracted features where we model them as a sequential input to the Transformer model. Furthermore, to improve the monotonicity correlation between the subjective and objective scores, we utilize the relative distance information among the images within each batch and enforce the relative ranking among them. Last but not least, we observe that the performance of NR-IQA models degrades when we apply equivariant transformations (e.g. horizontal flipping) to the inputs. Therefore, we propose a method that leverages self-consistency as a source of self-supervision to improve the robustness of NR-IQA models. Specifically, we enforce self-consistency between the outputs of our quality assessment model for each image and its transformation (horizontally flipped) to utilize the rich self-supervisory information and reduce the uncertainty of the model. To demonstrate the effectiveness of our work, we evaluate it on seven standard IQA datasets (both synthetic and authentic) and show that our model achieves state-of-the-art results on various datasets.1

The detection of spatially-varying blur without having any information about the blur type is a challenging task. In this paper, we propose a novel effective approach to address this blur detection problem from a single image without requiring any knowledge about the blur type, level, or camera settings. Our approach computes blur detection maps based on a novel High-frequency multiscale Fusion and Sort Transform (HiFST) of gradient magnitudes. The evaluations of the proposed approach on a diverse set of blurry images with different blur types, levels, and contents demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods qualitatively and quantitatively.

In this paper, we propose a reduced-reference (RR) objective quality assessment method that quantifies the perceived quality of synthesized textures. The proposed metric is based on measuring the granularity, regularity, and statistical attributes of the texture image. Furthermore, the proposed RR metric exhibits a significantly low overhead as compared to existing RR metrics by only requiring the transmission of 7 parameters as side information. Performance evaluations on two synthesized texture databases demonstrate that the proposed RR metric outperforms both full-reference (FR) and RR state-of-the-art quality metrics in predicting the perceived visual quality of the synthesized textures.

Perceptual image quality assessment (IQA) attempts to use computational models to estimate the image quality in accordance with subjective evaluations. Reduced-reference (RR) image quality assessment (IQA) methods make use of partial information or features extracted from the reference image for estimating the quality of distorted images. Finding a balance between the number of RR features and accuracy of the estimated image quality is essential and important in IQA. In this paper we propose a training-free low-cost RRIQA method that requires a very small number of RR features (6 RR features). The proposed RRIQA algorithm is based on the discrete wavelet transform (DWT) of locally weighted gradient magnitudes.We apply human visual system's contrast sensitivity and neighborhood gradient information to weight the gradient magnitudes in a locally adaptive manner. The RR features are computed by measuring the entropy of each DWT subband, for each scale, and pooling the subband entropies along all orientations, resulting in L RR features (one average entropy per scale) for an L-level DWT. Extensive experiments performed on seven large-scale benchmark databases demonstrate that the proposed RRIQA method delivers highly competitive performance as compared to the state-of-the-art RRIQA models as well as full reference ones for both natural and texture images. The MATLAB source code of REDLOG and the evaluation results are publicly available online at https://http://lab.engineering.asu.edu/ivulab/software/redlog/.

Reduced-reference (RR) image quality assessment (IQA) methods make use of partial information or features extracted from the reference image for estimating the quality of distorted images. Finding a balance between the number of RR features and image quality estimation accuracy is a difficult task. This paper presents a training-free low cost RRIQA method which requires a very small number of RR features (6 RR features). The proposed RRIQA algorithm is based on the divisive normalization transform (DNT) of locally weighted gradient magnitudes. The weighting of the gradient magnitudes is performed in a locally adaptive manner based on the human visual system's contrast sensitivity and neighborhood gradient information. The RR features are obtained by computing the entropy of each DNT subband and, for each scale, averaging the subband entropies along all orientations, resulting in L RR features (one average entropy per scale) for an L-level DNT. Performance evaluations on four large-scale benchmark databases demonstrates that the proposed RRIQA method delivers highly competitive performance as compared to the state-of-the-art RRIQA models as well as full reference ones.