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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]

EvFocus: Learning to Reconstruct Sharp Images from Out-of-Focus Event Streams

Authors: Lin Zhu, Xiantao Ma, Xiao Wang, Lizhi Wang, Hua Huang

ICML 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both simulated and real-world datasets demonstrate that Ev Focus outperforms existing methods across varying lighting conditions and blur sizes, proving its robustness and practical applicability in event-based defocus imaging.
Researcher Affiliation Academia 1School of Computer Science& Technology, Beijing Institute of Technology, Beijing, China 2School of Computer Science, Anhui University, Hefei, China 3School of Artificial Intelligence, Beijing Normal University, Beijing, China. Correspondence to: Hua Huang <EMAIL>.
Pseudocode Yes Algorithm 1 Synthetic Data Generation Pipeline Require: A set of background images {Bi} A set of foreground images {Fi,j} for each background i Motion parameters (e.g. translation, rotation) for background & foreground Number of time steps T Ensure: Synthetic dataset containing rendered scenes with events & optical flow 1: for each scene i do 2: Select one background image Bi 3: Select M foreground images {Fi,j}M j=1 4: Generate motion trajectories for background and each foreground: traj B Generate Trajectory(motion parameters) traj Fi,j Generate Trajectory(motion parameters) 5: for t = 1, . . . , T do 6: Sample camera pose pt Sample Pose() 7: Sample camera distortion dt Sample Distortion() 8: Render defocus brightness image: It Render(Bi, {Fi,j}, traj B[t], {traj Fi,j[t]}, pt, dt) 9: Compute brightness change It = It It 1 (if t > 1) 10: Generate events Et Event Generation( It) 11: Compute optical flow ut Optical Flow(It) 12: end for 13: Store {It}T t=1, {Et}T t=1, {ut}T t=1 as the dataset for scene i 14: end for
Open Source Code No The text does not contain any explicit statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets No As stated in Sec. 2, we generate sequences of defocus events, sharp images, and optical flows, in which 41 sequences are used in the training set and 6 sequences in the test set. To verify the effectiveness of our model on real data, we use the DAVIS 346 cameras to capture 7 real-world scenes.
Dataset Splits Yes As stated in Sec. 2, we generate sequences of defocus events, sharp images, and optical flows, in which 41 sequences are used in the training set and 6 sequences in the test set.
Hardware Specification Yes Our model is trained for 300 epochs with batch size of 1 on 3 NVIDIA Ge Force RTX 3090 GPUs.
Software Dependencies No Our model is implemented using the Py Torch framework. The paper mentions a software framework (PyTorch) but does not specify its version number or any other software dependencies with version information.
Experiment Setup Yes We adopt a constant strategy of learning rate during training, which is set at 1e-4. Our model is trained for 300 epochs with batch size of 1 on 3 NVIDIA Ge Force RTX 3090 GPUs.