Displacement-Invariant Matching Cost Learning for Accurate Optical Flow Estimation
Authors: Jianyuan Wang, Yiran Zhong, Yuchao Dai, Kaihao Zhang, Pan Ji, Hongdong Li
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that our approach achieves state-of-the-art accuracy on various datasets, and outperforms all published optical flow methods on the Sintel benchmark. |
| Researcher Affiliation | Collaboration | 1Australian National University, 2Northwestern Polytechnical University, 3NEC Labs America, 4Tencent AI Lab, 5ACRV |
| Pseudocode | No | The paper describes the method and uses equations, but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/jytime/DICL-Flow. |
| Open Datasets | Yes | The paper uses well-known public datasets: Flying Chair [5], Flying Things [21], Sintel [3], and KITTI [23], all with proper citations. |
| Dataset Splits | Yes | The paper mentions training on 'Flying Chair', 'Flying Things', 'Sintel', and 'KITTI' datasets, and refers to 'Flying Chair validation dataset' in Table 3's ablation study. This implies the use of standard, predefined splits for these well-known benchmarks. |
| Hardware Specification | Yes | The training details state: 'We use a batch size of 8 per GPU on the Flying Chair and 2 on other datasets, with 8 NVIDIA 1080 Ti GPUs for training.' |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions) needed to replicate the experiment. |
| Experiment Setup | Yes | The paper provides extensive experimental setup details including learning rates (0.001, 0.00025), training iterations (150K, 220K, 60K), learning rate drop schedules (half after 120K, at 30K, 50K), batch sizes (8, 2), loss function (multi-level ℓ2 loss with specific weights 1.0, 0.75, 0.5, 0.5, 0.5), and augmentation strategies (random resizing, cropping, flipping, color jittering, asymmetric occlusion). |