Context-Aware Iteration Policy Network for Efficient Optical Flow Estimation

Authors: Ri Cheng, Ruian He, Xuhao Jiang, Shili Zhou, Weimin Tan, Bo Yan

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that our method maintains performance while reducing FLOPs by about 40%/20% for the Sintel/KITTI datasets.
Researcher Affiliation Academia School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing, Fudan University rcheng22@m.fudan.edu.cn, rahe16@fudan.edu.cn, 20110240011@fudan.edu.cn, slzhou19@fudan.edu.cn, wmtan@fudan.edu.cn, byan@fudan.edu.cn
Pseudocode No The paper describes the method using equations and architectural diagrams (e.g., Figure 4) but does not include any formal pseudocode or algorithm blocks.
Open Source Code No The provided text does not contain an explicit statement about releasing open-source code for the described methodology or a link to a code repository.
Open Datasets Yes We measure average Endpoint Error (EPE) and the percentage of optical flow outliers over all pixels (F1-all) for Sintel (Butler et al. 2012) and KITTI (Menze and Geiger 2015) datasets. C+T refers to results that are trained on Chairs (Dosovitskiy et al. 2015) and Things (Mayer et al. 2016) datasets. S/K(+H) refers to methods fine-tuned on Sintel (Butler et al. 2012), KITTI (Menze and Geiger 2015), and some on HD1K (Kondermann et al. 2016) datasets.
Dataset Splits No The paper mentions 'training datasets' and 'test datasets' (Sintel, KITTI, Chairs, Things, HD1K) but does not specify explicit training/validation/test splits (e.g., percentages, sample counts, or specific predefined validation sets).
Hardware Specification Yes Each method was evaluated on an NVIDIA Ge Force RTX 3090 GPU to measure the inference speed per sample.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries (e.g., 'PyTorch 1.9', 'Python 3.8').
Experiment Setup Yes The weights λres and λincre in the overall loss (Equation 10) are set to 50 and 1. r is randomly sampled from 0.2 1.0. The learning rate is the same with their codes.