Multimodal Structure-Consistent Image-to-Image Translation
Authors: Che-Tsung Lin, Yen-Yi Wu, Po-Hao Hsu, Shang-Hong Lai11490-11498
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | For quantitative comparisons, we evaluate other competing methods and ours by using the generated images to train YOLO, Faster R-CNN and FCN models and prove that our model achieves significant improvement and outperforms other methods on the detection accuracies and the FCN scores. |
| Researcher Affiliation | Collaboration | Che-Tsung Lin,1,2 Yen-Yi Wu,1 Po-Hao Hsu,1 Shang-Hong Lai1,3 1Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan 2Intelligent Mobility Division, Mechanical and Mechatronics Systems Research Laboratories, Industrial Technology Research Institute, Hsinchu, Taiwan 3Microsoft AI R&D Center, Taipei, Taiwan |
| Pseudocode | No | The paper provides network architecture details and mathematical formulations for loss functions but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | No explicit statement or link regarding the availability of open-source code for the methodology described in the paper is provided. |
| Open Datasets | Yes | We conduct GAN model training and detector testing in SYNTHIA, GTA and BDD100k, respectively. ... SYNTHIA (Ros et al. 2016) and GTA (Richter et al. 2016), and real-driving datasets, such as BDD100k (Yu et al. 2018), provide detection and segmentation Ground-Truth in driving scenarios including different weather and time-of-day. |
| Dataset Splits | Yes | In GTA, all the daytime and the nighttime images in training sets are used in GAN training and the daytime images in validation set are transformed by GANs to train detectors which would be later assessed by the nighttime validation images. In BDD100k, the GAN training is done by using BDD100k-seg-train. The detectors are trained with day-to-night-transformed BDD100k-val-day and tested on BDD100k-val-night. |
| Hardware Specification | No | No specific hardware details (e.g., CPU, GPU models, memory, or cloud computing specifications) used for running the experiments are mentioned in the paper. |
| Software Dependencies | No | The paper states: "This work is implemented in Py Torch (Paszke et al. 2017)." However, it does not specify version numbers for PyTorch or any other software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | For training, we use the Adam optimizer (Kingma and Ba 2015) with a batch size of 4, a learning rate of 0.0002, exponential decay rates (β1, β2) = (0.5, 0.999). In all the experiments, we set the weightings related to structure consistency in the multi-task loss to be Lseg1 = Lseg2 = Lcycle1 = Lcycle2 = 5; others are all set to 1. |