SOTAVerified

Knowledge Distillation

Knowledge distillation is the process of transferring knowledge from a large model to a smaller one. While large models (such as very deep neural networks or ensembles of many models) have higher knowledge capacity than small models, this capacity might not be fully utilized.

Papers

Recently AddedMost HypedMost ActiveNeeds VerificationMost Verified

Showing 1–50 of 4240 papers

TitleStatusHype
O1 Replication Journey -- Part 2: Surpassing O1-preview through Simple Distillation, Big Progress or Bitter Lesson?Code7
AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-TimeCode5
A Survey on Knowledge Distillation of Large Language ModelsCode5
MobileSAMv2: Faster Segment Anything to EverythingCode5
Awesome Multi-modal Object TrackingCode5
Towards Cross-Tokenizer Distillation: the Universal Logit Distillation Loss for LLMsCode4
Vision-Language Models for Vision Tasks: A SurveyCode4
SAMPart3D: Segment Any Part in 3D ObjectsCode4
Effective Whole-body Pose Estimation with Two-stages DistillationCode4
Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo LabellingCode4
LLM Inference Unveiled: Survey and Roofline Model InsightsCode4
BitDistiller: Unleashing the Potential of Sub-4-Bit LLMs via Self-DistillationCode4
Generalized Robot 3D Vision-Language Model with Fast Rendering and Pre-Training Vision-Language AlignmentCode3
Logit Standardization in Knowledge DistillationCode3
Efficient Reasoning Models: A SurveyCode3
Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-DistillationCode3
PromptKD: Unsupervised Prompt Distillation for Vision-Language ModelsCode3
DistiLLM: Towards Streamlined Distillation for Large Language ModelsCode3
Efficient and Generalizable Speaker Diarization via Structured Pruning of Self-Supervised ModelsCode3
Compact Language Models via Pruning and Knowledge DistillationCode3
CMKD: CNN/Transformer-Based Cross-Model Knowledge Distillation for Audio ClassificationCode3
Recurrent Drafter for Fast Speculative Decoding in Large Language ModelsCode3
ProDiff: Progressive Fast Diffusion Model For High-Quality Text-to-SpeechCode3
N-LTP: An Open-source Neural Language Technology Platform for ChineseCode3
LLaVA-MoD: Making LLaVA Tiny via MoE Knowledge DistillationCode3
A Survey on Inference Optimization Techniques for Mixture of Experts ModelsCode3
Semi-Supervised Speech Recognition via Local Prior MatchingCode3
Focal Loss for Dense Object DetectionCode2
Anomaly Detection via Reverse Distillation from One-Class EmbeddingCode2
From Instance Training to Instruction Learning: Task Adapters Generation from InstructionsCode2
Event Stream-based Visual Object Tracking: A High-Resolution Benchmark Dataset and A Novel BaselineCode2
Event Stream-based Visual Object Tracking: HDETrack V2 and A High-Definition BenchmarkCode2
Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model InferenceCode2
Learning from Committee: Reasoning Distillation from a Mixture of Teachers with Peer-ReviewCode2
EPTQ: Enhanced Post-Training Quantization via Hessian-guided Network-wise OptimizationCode2
Efficient Large-scale Audio Tagging via Transformer-to-CNN Knowledge DistillationCode2
ECLIPSE: Efficient Continual Learning in Panoptic Segmentation with Visual Prompt TuningCode2
Efficient Multivariate Time Series Forecasting via Calibrated Language Models with Privileged Knowledge DistillationCode2
ESP-MedSAM: Efficient Self-Prompting SAM for Universal Domain-Generalized Medical Image SegmentationCode2
Distillation-Supervised Convolutional Low-Rank Adaptation for Efficient Image Super-ResolutionCode2
Diffusion Time-step Curriculum for One Image to 3D GenerationCode2
Decoupled Knowledge DistillationCode2
Distillation-Free One-Step Diffusion for Real-World Image Super-ResolutionCode2
DOT: A Distillation-Oriented TrainerCode2
ConDistFL: Conditional Distillation for Federated Learning from Partially Annotated DataCode2
OBSeg: Accurate and Fast Instance Segmentation Framework Using Segmentation Foundation Models with Oriented Bounding Box PromptsCode2
Cross-Image Relational Knowledge Distillation for Semantic SegmentationCode2
CoLaDa: A Collaborative Label Denoising Framework for Cross-lingual Named Entity RecognitionCode2
Data-Free Knowledge Distillation for Deep Neural NetworksCode2
Dual-Space Knowledge Distillation for Large Language ModelsCode2
Show:102550
← PrevPage 1 of 85Next →
All datasetsImageNetCIFAR-100COCO (Common Objects in Context)COCO 2017 valPASCAL VOCKITTI

Benchmark Results

#ModelMetricClaimedVerifiedStatus
1ScaleKD (T:BEiT-L S:ViT-B/14)Top-1 accuracy %86.43—Unverified
2ScaleKD (T:Swin-L S:ViT-B/16)Top-1 accuracy %85.53—Unverified
3ScaleKD (T:Swin-L S:ViT-S/16)Top-1 accuracy %83.93—Unverified
4ScaleKD (T:Swin-L S:Swin-T)Top-1 accuracy %83.8—Unverified
5KD++(T: regnety-16GF S:ViT-B)Top-1 accuracy %83.6—Unverified
6VkD (T:RegNety 160 S:DeiT-S)Top-1 accuracy %82.9—Unverified
7SpectralKD (T:Swin-S S:Swin-T)Top-1 accuracy %82.7—Unverified
8ScaleKD (T:Swin-L S:ResNet-50)Top-1 accuracy %82.55—Unverified
9DiffKD (T:Swin-L S: Swin-T)Top-1 accuracy %82.5—Unverified
10DIST (T: Swin-L S: Swin-T)Top-1 accuracy %82.3—Unverified
#ModelMetricClaimedVerifiedStatus
1SRD (T:resnet-32x4, S:shufflenet-v2)Top-1 Accuracy (%)79.86—Unverified
2shufflenet-v2(T:resnet-32x4, S:shufflenet-v2)Top-1 Accuracy (%)78.76—Unverified
3MV-MR (T: CLIP/ViT-B-16 S: resnet50)Top-1 Accuracy (%)78.6—Unverified
4resnet8x4 (T: resnet32x4 S: resnet8x4)Top-1 Accuracy (%)78.28—Unverified
5resnet8x4 (T: resnet32x4 S: resnet8x4 [modified])Top-1 Accuracy (%)78.08—Unverified
6ReviewKD++(T:resnet-32x4, S:shufflenet-v2)Top-1 Accuracy (%)77.93—Unverified
7ReviewKD++(T:resnet-32x4, S:shufflenet-v1)Top-1 Accuracy (%)77.68—Unverified
8resnet8x4 (T: resnet32x4 S: resnet8x4)Top-1 Accuracy (%)77.5—Unverified
9resnet8x4 (T: resnet32x4 S: resnet8x4)Top-1 Accuracy (%)76.68—Unverified
10resnet8x4 (T: resnet32x4 S: resnet8x4)Top-1 Accuracy (%)76.31—Unverified
#ModelMetricClaimedVerifiedStatus
1LSHFM (T: ResNet101 S: ResNet50)mAP77.16—Unverified
2LSHFM (T: ResNet101 S: MobileNetV2)mAP73.73—Unverified
3ADLIK-Faster (T: Faster R-CNN vit-base S: Faster R-CNN deit-small) box AP47.6—Unverified
4ADLIK-Mask (T: Mask R-CNN vit-base S: Mask R-CNN deit-small)mask AP42.4—Unverified
#ModelMetricClaimedVerifiedStatus
1ReviewKD++(T: faster rcnn(resnet101), S:faster rcnn(resnet50))AP@0.561.8—Unverified
2ReviewKD++(T: faster rcnn(resnet101), S:faster rcnn(resnet18))AP@0.557.96—Unverified
3ReviewKD++(T: faster rcnn(resnet101), S:faster rcnn(mobilenet-v2))AP@0.555.18—Unverified
#ModelMetricClaimedVerifiedStatus
1LSHFM (T: ResNet101 S: ResNet50)mAP93.17—Unverified
2LSHFM (T: ResNet101 S: MobileNetV2)mAP90.14—Unverified
#ModelMetricClaimedVerifiedStatus
1TIE-KD (T: Adabins S: MobileNetV2)RMSE2.43—Unverified