Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks88 from public data
- ByteTrack: Multi-object Tracking by Associating Every Detection Box2,841
A simple, effective and generic association method, tracking by associating almost every detection box instead of only the high score ones, named ByteTrack, which achieves state-of-the-art performance on MOT20, HiEve and BDD100K tracking benchmarks.
- Sparse R-CNN: End-to-End Object Detection with Learnable Proposals1,564
Sparse R-CNN demonstrates accuracy, run-time and training convergence performance on par with the well-established detector baselines on the challenging COCO dataset, e.g., achieving 45.0 AP in standard 3× training schedule and running at 22 fps using ResNet-50 FPN model.
- DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion467
A large-scale dataset for multi-human tracking, where humans have sim-ilar appearance, diverse motion and extreme articulation, and is named “DanceTrack”, which is expected to provide a better platform to develop more MOT algorithms that rely less on visual discrimination and depend more on motion analysis.
- Focal and Global Knowledge Distillation for Detectors432
Focal and Global Distillation (FGD) is proposed, which separates the foreground and background, forcing the student to focus on the teacher's critical pixels and channels, and compensating for missing global information in focal distillation.
- 345
- Infinity∞: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis335
In Infinity, a Bitwise Visual AutoRegressive Modeling capable of generating high-resolution, photorealistic images following language instruction, this work refactors visual autoregressive model under a bitwise token prediction framework with an infinite-vocabulary classifier and bit-wise self-correction mechanism.
- Universal Instance Perception as Object Discovery and Retrieval283
This work presents a universal instance perception model of the next generation, termed UNINEXT, which reformulates diverse instance perception tasks into a unified object discovery and retrieval paradigm and can flexibly perceive different types of objects by simply changing the input prompts.
- Language as Queries for Referring Video Object Segmentation282
This work proposes a simple and unified framework built upon Transformer, termed ReferFormer, which views the language as queries and directly attends to the most relevant regions in the video frames, and significantly outperforms the previous methods by a large margin.
- 268
- Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling162
This work identifies and overcome two key obstacles in extending the success of BERT-style pre-training, or the masked image modeling, to convolutional networks (convnets), and develops a hierarchical decoder to reconstruct images from multi-scale encoded features.
- Temporal Action Localization by Structured Maximal Sums155
This work addresses the problem of temporal action localization in videos by posing action localization as a structured prediction over arbitrary-length temporal windows, where each window is scored as the sum of frame-wise classification scores.
- 148
- Learning Object-Language Alignments for Open-Vocabulary Object Detection144
This paper proposes a novel open-vocabulary object detection framework directly learning from image-text pair data as a set matching problem between a set of image region features and aSet of word embeddings, which enables us to train an open- Vocabulary object detector on image- Text pairs in a much simple and effective way.
- 130
- General Object Foundation Model for Images and Videos at Scale120
GLEE is presented, an object-level foundation model for locating and identifying objects in images and videos, capable of being integrated into Large Language Models, serving as a foun-dational model to provide universal object-level information for multimodal tasks.
- 100
- 71
- 71
- Content-Variant Reference Image Quality Assessment via Knowledge Distillation60
The content-variant reference method via knowledge distillation (CVRKD-IQA) is proposed, which uses non-aligned reference (NAR) images to introduce various prior distributions of high-quality images and can support more IQA applications with its robustness to content variations.
- Deep-dense Conditional Random Fields for Object Co-segmentation59
A deep-dense conditional random field framework is introduced to infer co-occurrence maps of an image by combining two parts: objectness scores of image regions and similarity evidences from object proposals across images.
- Controllable Orthogonalization in Training DNNs48
This paper proposes a computationally efficient and numerically stable orthogonalization method using Newton's iteration (ONI), to learn a layer-wise Orthogonal weight matrix in DNNs and improves the performance of image classification networks by effectively controlling the orthogonality.
- 44
- Multimodal Transformer with Variable-Length Memory for Vision-and-Language Navigation44
This paper introduces Multimodal Transformer with Variable-length Memory (MTVM) for visually-grounded natural language navigation by modelling the temporal context explicitly and enables the agent to keep track of the navigation trajectory by directly storing previous activations in a memory bank.
- New Gradient-Spatial-Structural Features for video script identification40
This paper proposes to integrate the spatial and the structural features based on end points, intersection points, junction points and straightness of the skeleton of text components in a novel way to identify the scripts.
- Generative Region-Language Pretraining for Open-Ended Object Detection38
This paper forms object detection as a generative problem and proposes a simple framework named GenerateU, which can detect dense objects and generate their names in a free-form way and introduces an evaluation method designed to quantitatively measure the performance of generative out-comes.
- 35
- 35
- Self-supervised Video Representation Learning with Motion-Aware Masked Autoencoders35
Extensive experiments show that the MotionMAE outperforms significantly both supervised learning baseline and state-of-the-art MAE alternatives, under both domain-specific and domain-generic pretraining-then-finetuning settings.
- 35
- Multimodal Video Summarization via Time-Aware Transformers34
A Time-Aware Multimodal Transformer (TAMT) with a novel short-term order-sensitive attention mechanism that can attend the inputs differently based on time difference to explore the time information inherent inside video more thoroughly.
- 33
- Weakly Supervised Person Search with Region Siamese Networks33
This paper presents a weakly supervised setting where only bounding box annotations are available, and provides an effective baseline model termed Region Siamese Networks (R-SiamNets), which surpasses several fully supervised methods, such as OIM and MGTS, by a clear margin.
- 30
- Meta Compositional Referring Expression Segmentation28
This work proposes a Meta Compositional Referring Expression Segmentation (MCRES) framework, which can effectively drive the model to better capture semantics and visual representations of individual concepts, and thus obtain robust generalization performance even when handling novel compositions.
- 27
- 25
- 25
- Conditional Hyper-Network for Blind Super-Resolution With Multiple Degradations23
A novel conditional hyper-network framework for super-resolution with multiple degradations (named CMDSR) is proposed, which helps the SR framework learn how to adapt to changes in the degradation distribution of input and demonstrates the effectiveness ofCMDSR over various blind, and even several non-blind methods.
- 23
- 22
- 20
- 19
- 15
- 14
- 14
- 13
- 12
- 12
- 11
- 10
- 10
- 10
- 10
- 10
- 9
- 8
- Token Boosting for Robust Self-Supervised Visual Transformer Pre-training8
The Token Boosting Module (TBM) is introduced as a plug-and-play component for VTs that effectively allows the VT to learn to extract clean and robust features during masked autoencoding pre-training, and theoretical analysis is provided to show how TBM improves model pre- training with more robust and generalizable representations, thus benefiting down stream tasks.
- 8
- 7
- 5
- 4
- Single-Stage Open-world Instance Segmentation with Cross-task Consistency Regularization4
A regularization model to predict foreground pixels and use its relation to instance segmentation to construct a cross-task consistency loss that could alleviate the problem of incomplete instance annotation in Open-World Instance Segmentation.
- 3
- 3
- 3
- 3
- 3
- Transformer-based Open-world Instance Segmentation with Cross-task Consistency Regularization2
This work proposes a regularization model to predict foreground pixels and use its relation to instance segmentation to construct a cross-task consistency loss that could alleviate the problem of incomplete instance annotation in OWIS datasets.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-13. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.