In the pursuit of efficient automated content creation, procedural generation, leveraging modifiable parameters and rule-based systems, has emerged as a promising approach. Nonetheless, this can be a demanding endeavor, given its intricate nature necessitating a deep understanding of rules, algorithms, and parameters. To reduce workload, we introduce 3D-GPT, a framework utilizing large language models (LLMs) for instruction-driven 3D modeling. 3D-GPT positions LLMs as proficient problem solving agents, dissecting the procedural 3D modeling tasks into accessible segments and appointing the appropriate agent for each task. 3D-GPT integrates three core agents: the task dispatch agent, the conceptualization agent, and the modeling agent. They collaboratively achieve two objectives. First, they enhance the concise initial scene descriptions, elaborating details while dynamically adapting the text based on subsequent instructions. Second, they integrate procedural generation, extracting parameter values from enriched text to seamlessly interface with 3D software for asset creation. Our empirical investigations confirm that 3D-GPT not only correctly interprets instructions, delivering reliable results but also collaborates effectively with human designers. Furthermore, it integrates directly with Blender, unlocking expanded manipulation possibilities. Our work highlights the potential of LLMs in 3D modeling, offering a basic framework for future advancements in scene generation and animation.

Foundation models have significantly reduced the need for task-specific training, while also enhancing generaliz-ability. However, state-of-the-art 6D pose estimators either require further training with pose supervision or neglect advances obtainable from 3D foundation models. The latter is a missed opportunity, since these models are better equipped to predict 3D-consistent features, which are of significant utility for the pose estimation task. To address this gap, we propose Pos3R, a method for estimating the 6D pose of any object from a single RGB image, making extensive use of a 3D reconstruction foundation model and requiring no additional training. We identify template selection as a particular bottleneck for existing methods that is significantly alleviated by the use of a 3D model, which can more easily distinguish between template poses than a 2D model. Despite its simplicity, Pos3R achieves competitive performance on the Benchmark for 6D Object Pose Estimation (BOP), matching or surpassing existing refinement-free methods. Additionally, Pos3R integrates seamlessly with render-and-compare refinement techniques, demonstrating adaptability for high-precision applications.

Contemporary diffusion models built upon U-Net or Diffusion Transformer (DiT) architectures have revolutionized image generation through transformer-based attention mechanisms. The prevailing paradigm has commonly employed self-attention with quadratic computational complexity to handle global spatial relationships in complex images, thereby synthesizing high-fidelity images with coherent visual semantics.Contrary to conventional wisdom, our systematic layer-wise analysis reveals an interesting discrepancy: self-attention in pre-trained diffusion models predominantly exhibits localized attention patterns, closely resembling convolutional inductive biases. This suggests that global interactions in self-attention may be less critical than commonly assumed.Driven by this, we propose \(Δ\)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks (\(Δ\)ConvBlocks).By distilling attention patterns into localized convolutional operations while keeping other components frozen, \(Δ\)ConvFusion achieves performance comparable to transformer-based counterparts while reducing computational cost by 6929$\times$ and surpassing LinFusion by 5.42$\times$ in efficiency--all without compromising generative fidelity.

Assembling furniture amounts to solving the discrete-continuous optimization task of selecting the furniture parts to assemble and estimating their connecting poses in a physically realistic manner. The problem is hampered by its combinatorially large yet sparse solution space thus making learning to assemble a challenging task for current machine learning models. In this paper, we attempt to solve this task by leveraging the assembly instructions provided in diagrammatic manuals that typically accompany the furniture parts. Our key insight is to use the cues in these diagrams to split the problem into discrete and continuous phases. Specifically, we present Manual-PA, a transformer-based instruction Manual-guided 3D Part Assembly framework that learns to semantically align 3D parts with their illustrations in the manuals using a contrastive learning backbone towards predicting the assembly order and infers the 6D pose of each part via relating it to the final furniture depicted in the manual. To validate the efficacy of our method, we conduct experiments on the benchmark PartNet dataset. Our results show that using the diagrams and the order of the parts lead to significant improvements in assembly performance against the state of the art. Further, Manual-PA demonstrates strong generalization to real-world IKEA furniture assembly on the IKEA-Manual dataset.

Can the relative performance of a pre-trained large multimodal model (LMM) be predicted without access to labels? As LMMs proliferate, it becomes increasingly important to develop efficient ways to choose between them when faced with new data or tasks. The usual approach does the equivalent of giving the models an exam and marking them. We opt to avoid marking and the associated labor of determining the ground-truth answers. Instead, we explore other signals elicited and ascertain how well the models know their own limits, evaluating the effectiveness of these signals at unsupervised model ranking. We evaluate 47 state-of-the-art LMMs (e.g., LLaVA) across 9 visual question answering benchmarks, analyzing how well uncertainty-based metrics can predict relative model performance. Our findings show that uncertainty scores derived from softmax distributions provide a robust and consistent basis for ranking models across various tasks. This facilitates the ranking of LMMs on unlabeled data, providing a practical approach for selecting models for diverse target domains without requiring manual annotation.

To improve detection robustness in adverse conditions (e.g., haze and low light), image restoration is commonly applied as a pre-processing step to enhance image quality for the detector. However, the functional mismatch between restoration and detection networks can introduce instability and hinder effective integration---an issue that remains underexplored. We revisit this limitation through the lens of Lipschitz continuity, analyzing the functional differences between restoration and detection networks in both the input space and the parameter space. Our analysis shows that restoration networks perform smooth, continuous transformations, while object detectors operate with discontinuous decision boundaries, making them highly sensitive to minor perturbations. This mismatch introduces instability in traditional cascade frameworks, where even imperceptible noise from restoration is amplified during detection, disrupting gradient flow and hindering optimization. To address this, we propose Lipschitz-regularized object detection (LROD), a simple yet effective framework that integrates image restoration directly into the detector’s feature learning, harmonizing the Lipschitz continuity of both tasks during training. We implement this framework as Lipschitz-regularized YOLO (LR-YOLO), extending seamlessly to existing YOLO detectors. Extensive experiments on haze and low-light benchmarks demonstrate that LR-YOLO consistently improves detection stability, optimization smoothness, and overall accuracy.

Contrastive Language-Image Pre-training (CLIP) models have shown significant potential, particularly in zero-shot classification across diverse distribution shifts. Building on existing evaluations of overall classification robustness, this work aims to provide a more comprehensive assessment of CLIP by introducing several new perspectives. First, we investigate their robustness to variations in specific visual factors. Second, we assess two critical safety objectives-confidence uncertainty and out-of-distribution detection-beyond mere classification accuracy. Third, we evaluate the finesse with which CLIP models bridge the image and text modalities. Fourth, we extend our examination to 3D awareness in CLIP models, moving beyond traditional 2D image understanding. Finally, we explore the interaction between vision and language encoders within modern large multimodal models (LMMs) that utilize CLIP as the visual backbone, focusing on how this interaction impacts classification robustness. In each aspect, we consider the impact of six factors on CLIP models: model architecture, training distribution, training set size, fine-tuning, contrastive loss, and test-time prompts. Our study uncovers several previously unknown insights into CLIP. For instance, the architecture of the visual encoder in CLIP plays a significant role in their robustness against 3D corruption. CLIP models tend to exhibit a bias towards shape when making predictions. Moreover, this bias tends to diminish after fine-tuning on ImageNet. Vision-language models like LLaVA, leveraging the CLIP vision encoder, could exhibit benefits in classification performance for challenging categories over CLIP alone. Our findings are poised to offer valuable guidance for enhancing the robustness and reliability of CLIP models.

Neural implicit surface reconstruction leveraging volume rendering has led to significant advances in multi-view reconstruction. However, results for transparent objects can be very poor, primarily because the rendering function fails to account for the intricate light transport induced by refraction and reflection. In this study, we introduce trans-parent neural surface refinement (TNSR), a novel surface reconstruction framework that explicitly incorporates phys-ical refraction and reflection tracing. Beginning with an initial, approximate surface, our method employs sphere tracing combined with Snell's law to cast both reflected and refracted rays. Central to our proposal is an innovative differentiable technique devised to allow signals from the pho-tometric evidence to propagate back to the surface model by considering how the surface bends and reflects light rays. This allows us to connect surface refinement with volume rendering, enabling end-to-end optimization solely on multi-view RGB images. In our experiments, TNSR demonstrates significant improvements in novel view synthesis and geometry estimation of transparent objects, without prior knowledge of the refractive index.

Various deep-learning-based Low-dose Computed Tomography (LDCT) denoising methods have been proposed to address the issue of CT image noise, often relying on powerful network architecture and meticulously designed loss functions. However, these models are typically lacking efficient noise representations and prior anatomical knowledge, which can limit their overall performance. To address this issue, we introduce UNiversal Anatomy-initialized Noise Distribution Learning Framework (UNAD). The novelties of UNAD are in two aspects: 1) Distribution Regression Layer (DRL) restructures the output layer, enhancing noise prediction capability. 2) A simple yet efficient Predictive Pre-training Framework (PPT), tailored to the unique characteristics of CT images, is used to improve prior anatomical knowledge. We have achieved state-of-the-art results even without a powerful feature extractor like ViT. In addition, UNAD demonstrates remarkable universality, making it easy to incorporate into other LDCT denoising networks. The source code can be found in our official GitHub repository https://github.com/Nioolek/UNAD.

Vision-Language Models (VLMs) have emerged as the dominant approach for zero-shot recognition, adept at handling diverse scenarios and significant distribution changes. However, their deployment in risk-sensitive areas requires a deeper understanding of their uncertainty estimation capabilities, a relatively uncharted area. In this study, we explore the calibration properties of VLMs across different architectures, datasets, and training strategies. In particular, we analyze the uncertainty estimation performance of VLMs when calibrated in one domain, label set or hierarchy level, and tested in a different one. Our findings reveal that while VLMs are not inherently calibrated for uncertainty, temperature scaling significantly and consistently improves calibration, even across shifts in distribution and changes in label set. Moreover, VLMs can be calibrated with a very small set of examples. Through detailed experimentation, we highlight the potential applications and importance of our insights, aiming for more reliable and effective use of VLMs in critical, real-world scenarios.

Leveraging the model’s outputs, specifically the logits, is a common approach to estimating the test accuracy of a pre-trained neural network on out-of-distribution (OOD) samples without requiring access to the corresponding ground-truth labels. Despite their ease of implementation and computational efficiency, current logit-based methods are vulnerable to overconfidence issues, leading to prediction bias, especially under the natural shift. In this work, we first study the relationship between logits and generalization performance from the view of low-density separation assumption. Our findings motivate our proposed method \method{} that \textbf{(1)}~applies a data-dependent normalization on the logits to reduce prediction bias, and \textbf{(2)} takes the $L_p$ norm of the matrix of normalized logits as the estimation score. Our theoretical analysis highlights the connection between the provided score and the model's uncertainty. We conduct an extensive empirical study on common unsupervised accuracy estimation benchmarks and demonstrate that \method{} achieves state-of-the-art performance across various architectures in the presence of synthetic, natural, or subpopulation shifts. The code is available at https://github.com/Renchunzi-Xie/MaNo.

Neural Radiance Fields (NeRF) have demonstrated exceptional capabilities in creating photorealistic novel views using volume rendering on a radiance field. However, the intrinsic assumption of straight light rays within NeRF becomes a limitation when dealing with transparent or translucent objects that exhibit refraction, and therefore have curved light paths. This hampers the ability of these approaches to accurately model the appearance of refractive objects, resulting in suboptimal novel view synthesis and geometry estimates. To address this issue, we propose an innovative solution using deformable networks to learn a tailored deformation field for refractive objects. Our approach predicts position and direction offsets, allowing NeRF to model the curved light paths caused by refraction and therefore the complex and highly view-dependent appearances of refractive objects. We also introduce a regularization strategy that encourages piece-wise linear light paths, since most physical systems can be approximated with a piece-wise constant index of refraction. By seamlessly integrating our deformation networks into the NeRF framework, our method significantly improves rendering refractive objects from novel views.

Understanding model decision under novel test scenarios is central to the community. A common practice is evaluating models on labeled test sets. However, many real-world scenarios see unlabeled test data, rendering the common supervised evaluation protocols infeasible. In this paper, we investigate such an important but under-explored problem, named Automatic model Evaluation (AutoEval). Specifically, given a trained classifier, we aim to estimate its accuracy on various unlabeled test datasets. We construct a meta-dataset: a dataset comprised of datasets (sample sets) created from original images via various transformations such as rotation and background substitution. Correlation studies on the meta-dataset show that classifier accuracy exhibits a strong negative linear relationship with distribution shift (Pearson's Correlation ). This new finding inspires us to formulate AutoEval as a dataset-level regression problem. Specifically, we learn regression models (e.g., a regression neural network) to estimate classifier accuracy from overall feature statistics of a test set. In the experiment, we show that the meta-dataset contains sufficient and diverse sample sets, allowing us to train robust regression models and report reasonable and promising predictions of the classifier accuracy on various test sets. We also provide insights into application scopes, limitations, and potential future directions of AutoEval.

This work investigates dataset vectorization for two dataset-level tasks: assessing training set suitability and test set difficulty. The former measures how suitable a training set is for a target domain, while the latter studies how challenging a test set is for a learned model. Central to the two tasks is measuring the underlying relationship between datasets. This needs a desirable dataset vectorization scheme, which should preserve as much discriminative dataset information as possible so that the distance between the resulting dataset vectors can reflect dataset-to-dataset similarity. To this end, we propose a bag-of-prototypes (BoP) dataset representation that extends the image-level bag consisting of patch descriptors to dataset-level bag consisting of semantic prototypes. Specifically, we develop a codebook consisting of K prototypes clustered from a ref-erence dataset. Given a dataset to be encoded, we quan-tize each of its image features to a certain prototype in the codebook and obtain a K -dimensional histogram. Without assuming access to dataset labels, the BoP representation provides rich characterization of the dataset semantic distribution. Furthermore, BoP representations cooperate well with Jensen-Shannon divergence for measuring dataset-to-dataset similarity. Although very simple, BoP consistently shows its advantage over existing representations on a series of benchmarks for two dataset-level tasks.

Model calibration usually requires optimizing some parameters (e.g., temperature) w.r.t an objective function like negative log-likelihood. This work uncovers a significant but overlooked aspect that the objective function is influenced by calibration set difficulty: the ratio of misclassified to correctly classified samples1. If a test set has a drastically different difficulty level from the calibration set, a phenomenon out-of-distribution (OOD) data often exhibit: the optimal calibration parameters of the two datasets would be different, rendering an optimal calibrator on the calibration set suboptimal on the OOD test set and thus degraded calibration performance. With this knowledge, we propose a simple and effective method named adaptive calibrator ensemble (ACE) to calibrate OOD datasets whose difficulty is usually higher than the calibration set. Specifically, two calibration functions are trained, one for in-distribution data (low difficulty), and the other for severely OOD data (high difficulty). To achieve desirable calibration on a new OOD dataset, ACE uses an adaptive weighting method that strikes a balance between the two extreme functions. When plugged in, ACE generally improves the performance of a few state-of-the-art calibration schemes on a series of OOD benchmarks. Importantly, such improvement does not come at the cost of the in-distribution calibration performance. Project Website: https://github.com/insysgroup/Adaptive-Calibrators-Ensemble.git.

This work aims to assess how well a model performs under distribution shifts without using labels. While recent methods study prediction confidence, this work reports prediction dispersity is another informative cue. Confidence reflects whether the individual prediction is certain; dispersity indicates how the overall predictions are distributed across all categories. Our key insight is that a well-performing model should give predictions with high confidence and high dispersity. That is, we need to consider both properties so as to make more accurate estimates. To this end, we use nuclear norm that has been shown to be effective in characterizing both properties. Extensive experiments validate the effectiveness of nuclear norm for various models (e.g., ViT and ConvNeXt), different datasets (e.g., ImageNet and CUB-200), and diverse types of distribution shifts (e.g., style shift and reproduction shift). We show that nuclear norm is more accurate and robust in accuracy estimation than existing methods. Furthermore, we validate the feasibility of other measurements (e.g., mutual information maximization) for characterizing dispersity and confidence. Lastly, we investigate the limitation of the nuclear norm, study its improved variant under severe class imbalance, and discuss potential directions.

Contrastive Language-Image Pre-training (CLIP) models have demonstrated remarkable generalization capabilities across multiple challenging distribution shifts. However, there is still much to be explored in terms of their robustness to the variations of specific visual factors. In real-world applications, reliable and safe systems must consider other safety measures beyond classification accuracy, such as predictive uncertainty. Yet, the effectiveness of CLIP models on such safety-related objectives is less-explored. Driven by the above, this work comprehensively investigates the safety measures of CLIP models, specifically focusing on three key properties: resilience to visual factor variations, calibrated uncertainty estimations, and the ability to detect anomalous inputs. To this end, we study $83$ CLIP models and $127$ ImageNet classifiers. They are diverse in architecture (pre)training distribution and training strategies. We consider $10$ visual factors (\emph{e.g.}, shape and pattern), $5$ types of out-of-distribution data, and $8$ natural and challenging test conditions with different shift types, such as texture, style, and perturbation shifts. Our study has unveiled several previously unknown insights into CLIP models. For instance, they are not consistently more calibrated than other ImageNet models, which contradicts existing findings. Additionally, our analysis underscores the significance of training source design by showcasing its profound influence on the three key properties. We believe our comprehensive study can shed light on and help guide the development of more robust and reliable CLIP models.

This paper presents an approach to train a unified deep network that simultaneously solves multiple human-related tasks. A multi-task framework is favorable for sharing information across tasks under restricted computational resources. However, tasks not only share information but may also compete for resources and conflict with each other, making the optimization of shared parameters difficult and leading to suboptimal performance. We propose a simple but effective training scheme called GradSplit that alleviates this issue by utilizing asymmetric inter-task relations. Specifically, at each convolution module, it splits features into T groups for T tasks and trains each group only using the gradient back-propagated from the task losses with which it does not have conflicts. During training, we apply GradSplit to a series of convolution modules. As a result, each module is trained to generate a set of task-specific features using the shared features from the previous module. This enables a network to use complementary information across tasks while circumventing gradient conflicts. Experimental results show that GradSplit achieves a better accuracy-efficiency trade-off than existing methods. It minimizes accuracy drop caused by task conflicts while significantly saving compute resources in terms of both FLOPs and memory at inference. We further show that GradSplit achieves higher cross-dataset accuracy compared to single-task and other multi-task networks.

Generalization and invariance are two essential properties of machine learning models. Generalization captures a model's ability to classify unseen data while invariance measures consistency of model predictions on transformations of the data. Existing research suggests a positive relationship: a model generalizing well should be invariant to certain visual factors. Building on this qualitative implication we make two contributions. First, we introduce effective invariance (EI), a simple and reasonable measure of model invariance which does not rely on image labels. Given predictions on a test image and its transformed version, EI measures how well the predictions agree and with what level of confidence. Second, using invariance scores computed by EI, we perform large-scale quantitative correlation studies between generalization and invariance, focusing on rotation and grayscale transformations. From a model-centric view, we observe generalization and invariance of different models exhibit a strong linear relationship, on both in-distribution and out-of-distribution datasets. From a dataset-centric view, we find a certain model's accuracy and invariance linearly correlated on different test sets. Apart from these major findings, other minor but interesting insights are also discussed.

Motivated by the desire to exploit patterns shared across classes, we present a simple yet effective class-specific memory module for fine-grained feature learning. The memory module stores the prototypical feature representation for each category as a moving average. We hypothesize that the combination of similarities with respect to each category is itself a useful discriminative cue. To detect these similarities, we use attention as a querying mechanism. The attention scores with respect to each class prototype are used as weights to combine prototypes via weighted sum, producing a uniquely tailored response feature representation for a given input. The original and response features are combined to produce an augmented feature for classification. We integrate our class-specific memory module into a standard convolutional neural network, yielding a Categorical Memory Network. Our memory module significantly improves accuracy over baseline CNNs, achieving competitive accuracy with state-of-the-art methods on four benchmarks, including CUB-200-2011, Stanford Cars, FGVC Aircraft, and NABirds.

To calculate the model accuracy on a computer vision task, e.g., object recognition, we usually require a test set composing of test samples and their ground truth labels. Whilst standard usage cases satisfy this requirement, many real-world scenarios involve unlabeled test data, rendering common model evaluation methods infeasible. We investigate this important and under-explored problem, Automatic model Evaluation (AutoEval). Specifically, given a labeled training set and a classifier, we aim to estimate the classification accuracy on unlabeled test datasets. We construct a meta-dataset: a dataset comprised of datasets generated from the original images via various transformations such as rotation, background substitution, foreground scaling, etc. As the classification accuracy of the model on each sample (dataset) is known from the original dataset labels, our task can be solved via regression. Using the feature statistics to represent the distribution of a sample dataset, we can train regression models (e.g., a regression neural network) to predict model performance. Using synthetic meta-dataset and real-world datasets in training and testing, respectively, we report a reasonable and promising prediction of the model accuracy. We also provide insights into the application scope, limitation, and potential future direction of AutoEval.

Consider a scenario where we are supplied with a number of ready-to-use models trained on a certain source domain and hope to directly apply the most appropriate ones to different target domains based on the models’ relative performance. Ideally we should annotate a validation set for model performance assessment on each new target environment, but such annotations are often very expensive. Under this circumstance, we introduce the problem of ranking models in unlabeled new environments. For this problem, we propose to adopt a proxy dataset that 1) is fully labeled and 2) well reflects the true model rankings in a given target environment, and use the performance rankings on the proxy sets as surrogates. We first select labeled datasets as the proxy. Specifically, datasets that are more similar to the unlabeled target domain are found to better preserve the relative performance rankings. Motivated by this, we further propose to search the proxy set by sampling images from various datasets that have similar distributions as the target. We analyze the problem and its solutions on the person re-identification (re-ID) task, for which sufficient datasets are publicly available, and show that a carefully constructed proxy set effectively captures relative performance ranking in new environments. Code is avalible at https://github.com/sxzrt/Proxy-Set.

Understanding classifier decision under novel environments is central to the community, and a common practice is evaluating it on labeled test sets. However, in real-world testing, image annotations are difficult and expensive to obtain, especially when the test environment is changing. A natural question then arises: given a trained classifier, can we evaluate its accuracy on varying unlabeled test sets? In this work, we train semantic classification and rotation prediction in a multi-task way. On a series of datasets, we report an interesting finding, i.e., the semantic classification accuracy exhibits a strong linear relationship with the accuracy of the rotation prediction task (Pearson’s Correlation r > 0.88). This finding allows us to utilize linear regression to estimate classifier performance from the accuracy of rotation prediction which can be obtained on the test set through the freely generated rotation labels.

The gap in data distribution motivates domain adaptation research. In this area, image classification intrinsically requires the source and target features to be co-located if they are of the same class. However, many works only take a global view of the domain gap. That is, to make the data distributions globally overlap; and this does not necessarily lead to feature co-location at the class level. To resolve this problem, we study metric learning in the context of domain adaptation. Specifically, we introduce a similarity guided constraint (SGC). In the implementation, SGC takes the form of a triplet loss. The triplet loss is integrated into the network as an additional objective term. Here, an image triplet consists of two images of the same class and another image of a different class. Albeit simple, the working mechanism of our method is interesting and insightful. Importantly, images in the triplets are sampled from the source and target domains. From a micro perspective, by enforcing this constraint on every possible triplet, images from different domains but of the same class are mapped nearby, and those of different classes are far apart. From a macro perspective, our method ensures that cross-domain similarities are preserved, leading to intra-class compactness and inter-class separability. Extensive experiment on four datasets shows our method yields significant improvement over the baselines and has a competitive accuracy with the state-of-the-art results.

Person re-identification (re-ID) models trained on one domain often fail to generalize well to another. In our attempt, we present a "learning via translation" framework. In the baseline, we translate the labeled images from source to target domain in an unsupervised manner. We then train re-ID models with the translated images by supervised methods. Yet, being an essential part of this framework, unsupervised image-image translation suffers from the information loss of source-domain labels during translation. Our motivation is two-fold. First, for each image, the discriminative cues contained in its ID label should be maintained after translation. Second, given the fact that two domains have entirely different persons, a translated image should be dissimilar to any of the target IDs. To this end, we propose to preserve two types of unsupervised similarities, 1) self-similarity of an image before and after translation, and 2) domain-dissimilarity of a translated source image and a target image. Both constraints are implemented in the similarity preserving generative adversarial network (SPGAN) which consists of an Siamese network and a CycleGAN. Through domain adaptation experiment, we show that images generated by SPGAN are more suitable for domain adaptation and yield consistent and competitive re-ID accuracy on two large-scale datasets.

This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (reID). We view each weight vector within a fully connected (FC) layer in a convolutional neuron network (CNN) as a projection basis. It is observed that the weight vectors are usually highly correlated. This problem leads to correlations among entries of the FC descriptor, and compromises the retrieval performance based on the Euclidean distance. To address the problem, this paper proposes to optimize the deep representation learning process with Singular Vector Decomposition (SVD). Specifically, with the restraint and relaxation iteration (RRI) training scheme, we are able to iteratively integrate the orthogonality constraint in CNN training, yielding the so-called SVDNet. We conduct experiments on the Market-1501, CUHK03, and DukeMTMC-reID datasets, and show that RRI effectively reduces the correlation among the projection vectors, produces more discriminative FC descriptors, and significantly improves the re-ID accuracy. On the Market-1501 dataset, for instance, rank-1 accuracy is improved from 55.3% to 80.5% for CaffeNet, and from 73.8% to 82.3% for ResNet-50.