Résultats de recherche pour « Co-DETR »

Automated analysis of Pap smear images is critical for cervical cancer screening but remains challenging due to dense cell distribution and complex morphology. In this paper, we present our winning solution for the RIVA Cervical Cytology Challenge, achieving 1st place in Track B and 2nd place in Track A. Our approach leverages a powerful baseline, integrating the Co-DINO framework with a Swin-Large backbone for robust multi-scale feature extraction. To address the dataset's unique fixed-size bounding box annotations, we formulate the detection task as a center-point prediction problem. Tailoring our approach to this formulation, we introduce a center-preserving data augmentation strategy and an analytical geometric box optimization to effectively absorb localization jitter. Finally, we apply track-specific loss tuning to adapt the loss weights for each task. Experiments demonstrate that our targeted optimizations improve detection performance, providing an effective pipeline for cytology image analysis. Our code is available at https://github.com/YanKong0408/Center-DETR.

Employing computer vision-based models for intelligent pipeline defect detection offers a promising alternative to the conventional labor-intensive and time-consuming interpretation of CCTV videos. To better improve the capabilities of state-of-the-art object detection models addressing complex subterranean environment, this paper proposes a pipeline inspection model, P-CoDETR, with an enhanced Co-DETR architecture for defect detections under restricted field of view with imbalanced sample images. This model incorporates a Relation-Aware Self-Attention (RASA) mechanism within its backbone network, establishing an RSwinTransformer architecture to enhance modeling capabilities for complex geometries and spatial relationships. Concurrently, an Adaptive Threshold Focal Loss (ATFL) function is designed to dynamically adjust and focus gradients on challenging samples, thereby improving robustness and discrimination. Validation experiments have been conducted using a self-prepared large-scale dataset of 10,004 pipeline defect images of nine categories. The results demonstrate that compared to the original Co-DETR and mainstream YOLO/DETR series models, P-CoDETR achieves significant improvements in mAP and precision across defect types and model stability, exhibiting promising application prospects in replacing experienced engineers in pipeline automatic assessments. Notably, the average precision of total nine defect categories reaches 91.2% with remarkable improvements in detecting geometrically sensitive defects, such as deposition, corrosion and encrustation.

The recently proposed Detection Transformer (DETR) model successfully applies Transformer to objects detection and achieves comparable performance with two-stage object detection frameworks, such as Faster-RCNN. However, DETR suffers from its slow convergence. Training DETR from scratch needs 500 epochs to achieve a high accuracy. To accelerate its convergence, we propose a simple yet effective scheme for improving the DETR framework, namely Spatially Modulated Co-Attention (SMCA) mechanism. The core idea of SMCA is to conduct location-aware co-attention in DETR by constraining co-attention responses to be high near initially estimated bounding box locations. Our proposed SMCA increases DETR's convergence speed by replacing the original co-attention mechanism in the decoder while keeping other operations in DETR unchanged. Furthermore, by integrating multi-head and scale-selection attention designs into SMCA, our fully-fledged SMCA can achieve better performance compared to DETR with a dilated convolution-based backbone (45.6 mAP at 108 epochs vs. 43.3 mAP at 500 epochs). We perform extensive ablation studies on COCO dataset to validate SMCA. Code is released at https://github.com/gaopengcuhk/SMCA-DETR .

The recently proposed Detection Transformer (DETR) model successfully applies Transformer to objects detection and achieves comparable performance with two-stage object detection frameworks, such as Faster-RCNN. However, DETR suffers from its slow convergence. Training DETR \cite{carion2020end} from scratch needs 500 epochs to achieve a high accuracy. To accelerate its convergence, we propose a simple yet effective scheme for improving the DETR framework, namely Spatially Modulated Co-Attention (SMCA) mechanism. The core idea of SMCA is to conduct regression-aware co-attention in DETR by constraining co-attention responses to be high near initially estimated bounding box locations. Our proposed SMCA increases DETR's convergence speed by replacing the original co-attention mechanism in the decoder while keeping other operations in DETR unchanged. Furthermore, by integrating multi-head and scale-selection attention designs into SMCA, our fully-fledged SMCA can achieve better performance compared to DETR with a dilated convolution-based backbone (45.6 mAP at 108 epochs vs. 43.3 mAP at 500 epochs). We perform extensive ablation studies on COCO dataset to validate the effectiveness of the proposed SMCA.

This code is the supporting research code for the paper "Weight Activation Co Adaptive Quantization Method for RT-DETR", which implements the weight activation co adaptive quantization (WCAQ) framework for studying the post training quantization of real-time object detection model RT-DETR under low bit conditions. This framework combines group weight pseudo quantization, sample adaptive activation scale adjustment, operator output activation quantization, sensitive branch protection, and layer by layer reconstruction to alleviate the degradation of detection accuracy caused by joint quantization of weights and activations without the need for task level retraining. The code covers model quantization configuration, calibration, layer by layer reconstruction, and detection performance evaluation. It supports conducting 8-bit, 6-bit, and 4-bit quantization experiments on the COCO 2017 dataset and can be used for paper method reproduction and related model quantization research. The current implementation uses pseudo quantization to simulate low bit numerical calculations, and some sensitive modules retain floating-point operations, which is not yet a fully networked pure integer inference implementation.

In this paper, we provide the observation that too few queries assigned as positive samples in DETR with one-to-one set matching leads to sparse supervision on the encoder’s output which considerably hurt the discriminative feature learning of the encoder and vice visa for attention learning in the decoder. To alleviate this, we present a novel collaborative hybrid assignments training scheme, namely $\mathcal{C}o - {\text{DETR}}$, to learn more efficient and effective DETR-based detectors from versatile label assignment manners. This new training scheme can easily enhance the encoder’s learning ability in end-to-end detectors by training the multiple parallel auxiliary heads supervised by one-to-many label assignments such as ATSS and Faster RCNN. In addition, we conduct extra customized positive queries by extracting the positive coordinates from these auxiliary heads to improve the training efficiency of positive samples in the decoder. In inference, these auxiliary heads are discarded and thus our method introduces no additional parameters and computational cost to the original detector while requiring no hand-crafted non-maximum suppression (NMS). We conduct extensive experiments to evaluate the effectiveness of the proposed approach on DETR variants, including DAB-DETR, Deformable-DETR, and DINO-Deformable-DETR. The state-of-the-art DINO-Deformable-DETR with Swin-L can be improved from 58.5% to 59.5% AP on COCO val. Surprisingly, incorporated with ViT-L backbone, we achieve 66.0% AP on COCO test-dev and 67.9% AP on LVIS val, outperforming previous methods by clear margins with much fewer model sizes. Codes are available at https://github.com/Sense-X/Co-DETR.

In this study, an Infrared Small-Target Detection Transformer (IST-DETR) is introduced, a novel model specifically designed to tackle the challenges of low-resolution infrared images and small target scales. IST-DETR integrates a backbone network, a hybrid encoder featuring, a Mutual Feature Screening (MFS) module, and a decoder with auxiliary prediction heads. The hybrid encoder employs Learning Position Encoding to reduce information redundancy and employs a mutual feature screening mechanism to enhance the interaction between high-level semantic features and low-level positional features, facilitating more accurate detection of small infrared targets. What’s more, a customized IoU metric and a novel sample weighting function are employed to effectively address dataset imbalance, significantly improving detection performance. Experiments conducted on the FLIR Dataset, HIT-UAV Dataset, and IVFlying Dataset yielded an average precision (AP) of 44.1%, 34.0%, and 58.5%, respectively, with a processing speed of 74 frames per second. IST-DETR outperforms contemporary algorithms such as Yolov8, CO-DETR, and DINO, demonstrating a superior balance of speed and accuracy, particularly in recognizing small infrared targets across diverse and complex scenarios.

Otitis media (OM) and its complications can cause immense suffering for patients. However, due to the limited number of experts and their knowledge base, human specialists can only handle a limited number of CT images. Therefore, the auxiliary diagnosis of OM is crucial. Object detection is one of the primary tasks in computer vision. With the development of Transformers and attention mechanisms in recent years, the DETR series detectors have gradually become mainstream. However, these detectors are highly sensitive to the number of decoder layers, and slight deviations in the layers significantly impair model performance. The required number of decoder layers varies across different datasets, with grayscale images such as CT scans likely needing a different number of decoder layers than color images. Considering the stringent requirements for performance in clinical CT diagnostics, this paper analyzes the decline in model effectiveness due to excessive decoder layers. It proposes a new model called 4DO-DETR to address the instability issues of DETR performance. Regarding performance, 4DO-DETR has shown significant improvement over the baseline model (DN-DAB-DETR), surpassing the state-of-the-art (SOTA) algorithms Co-DETR, DINO and RT-DETR in the past two years. Rigorous experimental evaluations on a benchmark medical dataset demonstrate that our method achieves higher scores than other sophisticated network models——our model’s mAP reached 56.8%, higher than DINO’s 54.7%, Co-DETR’s 54.0%, and the baseline’s 45.1%. The datasets employed in this paper are available at https://github.com/promisedong/Four-DO-DETR .

Automatic Vehicle Detection (AVD) in diverse driving environments presents unique challenges due to varying lighting conditions, road types, and vehicle types. Traditional methods, such as YOLO and Faster R-CNN, often struggle to cope with these complexities. As computer vision evolves, combining Convolutional Neural Networks (CNNs) with Transformer-based approaches offers promising opportunities for improving detection accuracy and efficiency. This study is the first to experiment with Detection Transformer (DETR) for automatic vehicle detection in complex and varied settings. We employ a Collaborative Hybrid Assignments Training scheme, Co-DETR, to enhance feature learning and attention mechanisms in DETR. By leveraging versatile label assignment strategies and introducing multiple parallel auxiliary heads, we provide more effective supervision during training and extract positive coordinates to boost training efficiency. Through extensive experiments on DETR variants and YOLO models, conducted using the BadODD dataset, we demonstrate the advantages of our approach. Our method achieves superior results, and improved accuracy in diverse conditions, making it practical for real-world deployment. This work significantly advances autonomous navigation technology and opens new research avenues in object detection for autonomous vehicles. By integrating the strengths of CNNs and Transformers, we highlight the potential of DETR for robust and efficient vehicle detection in challenging driving environments.