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..
Polyline Path Masked Attention for Vision Transformer
Authors: Zhongchen Zhao, Chaodong Xiao, Hui LIN, Qi Xie, Lei Zhang, Deyu Meng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on standard benchmarks, including image classification, object detection, and segmentation, demonstrate that our model outperforms previous state-of-the-art approaches based on both state-space models and Transformers. |
| Researcher Affiliation | Collaboration | Zhongchen Zhao1, Chaodong Xiao2,3, Hui Lin1, Qi Xie1, , Lei Zhang2,3, Deyu Meng1,4 1Xi an Jiaotong University 2The Hong Kong Polytechnic University 3OPPO Research 4Pazhou Laboratory (Huangpu) Institute |
| Pseudocode | Yes | Algorithm 1: Efficient Masked Attention Computation. ... Algorithm 2: Efficient Masked Attention Computation. |
| Open Source Code | Yes | Code is available at https://github.com/zhongchenzhao/PPMA. |
| Open Datasets | Yes | We evaluate the classification performance of our method on Image Net-1K [7]... object detection and instance segmentation tasks on MSCOCO2017 [28]... semantic segmentation performance of our method on ADE20K [56] |
| Dataset Splits | Yes | We evaluate the classification performance of our method on Image Net-1K [7]... on COCO val2017... on ADE20K val set. |
| Hardware Specification | Yes | To evaluate the inference speed of our model, we measure the throughput of PPMA-T/S/B on an A800 GPU with a batch size of 64 and the image resolution of 224 224. |
| Software Dependencies | No | For experiments on the ADE20K [56] and MSCOCO2017 [28] datasets, we follow the training settings of Trans Ne XT [35], and utilize the MMDetection [2] and MMSegmentation [4] libraries for training. |
| Experiment Setup | Yes | Following the same training strategy as in [10, 44], we train our models from scratch for 300 epochs with the input size of 224 224. We use the adaptive Adam W optimizer with a cosine decay learning rate scheduler (batch size=1024, initial learning rate=0.001, weight decay=0.05). |