Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
ELDET: Early-Learning Distillation with Noisy Labels for Object Detection
Authors: Dongmin Choi, Sangbin Lee, EungGu Yun, Jonghyuk Baek, Frank C. Park
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach on PASCAL VOC and MS COCO with noise simulations. Additionally, we assess the domain robustness of our framework on the Vin Dr-CXR medical detection dataset [39], demonstrating its adaptability to specialized domains. Extensive experimental results demonstrate that our method significantly improves detection performance in noisy environments |
| Researcher Affiliation | Collaboration | Dongmin Choi1* Sangbin Lee1,2* Eung Gu Yun1 Jonghyuk Baek3 Frank C. Park1,2 1SAIGE 2Seoul National University 3Flitto EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the method in prose and through mathematical equations (e.g., Equation 1, 2, 3, 4, 5, 6, 7, 8, 9) but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | Answer: [No] Justification: We intend to release anonymized code with the camera-ready version. At submission time, the code is not included in the supplementary material. |
| Open Datasets | Yes | We evaluate our approach on PASCAL VOC and MS COCO with noise simulations. Additionally, we assess the domain robustness of our framework on the Vin Dr-CXR medical detection dataset [39]... We use the COCO 2017 version... Vin Dr-CXR dataset [40]... |
| Dataset Splits | Yes | Following the standard protocol [49], we use the VOC 2007 and VOC 2012 trainval sets (16,551 images) for training, and perform evaluation on the VOC 2007 test set (4,952 images). MS COCO is a large-scale dataset with 80 object categories, featuring over 330K images and more than 2.5 million labeled instances. We use the COCO 2017 version, training on the train split (118K images) and evaluating on the val split (5K images). |
| Hardware Specification | Yes | All experiments were conducted using GPUs with 24GB VRAM (NVIDIA RTX 3090 and 4090). |
| Software Dependencies | No | Our proposed method is implemented using the MMDetection framework [9] built on Py Torch [42]. |
| Experiment Setup | Yes | Training is conducted using Stochastic Gradient Descent (SGD) with a momentum of 0.9 and a weight decay of 10 4. The learning rate follows a step schedule, decreasing by a factor of 10 at predefined epochs, except for MS COCO [30] where only a linear scheduler is used. For PASCAL VOC [11] and MS COCO, the learning rate is set to 0.01, and the training spans 12 epochs with a batch size of 32. In contrast, for Vin Dr-CXR [40], the learning rate is set to 0.005, and the training spans 20 epochs with a batch size of 16. |