Home

Awesome

Implicit Neural Image Stitching With Enhanced and Blended Feature Reconstruction

This repository contains the official implementation of NIS, 24' WACV: https://arxiv.org/abs/2309.01409

Requirement

  1. Python packages
conda env create --file environment.yaml
conda activate nis
  1. pysrwarp : Follow the guidelines in the repository for more details and compile debugging.
git clone https://github.com/sanghyun-son/pysrwarp
cd pysrwarp
make

Dataset

Pretrained Models

You can download below models on this link.

  1. NIS_enhancing.pth: Pretrained on Enhanced Stitching (Stage 1),
  2. NIS_blending.pth: Pretrained on Enhanced & Blended Stitching (Stage 1 & 2),
  3. ihn.pth: Our reproduced Homography Estimator used in the second stage training.

Train & Evaluation

bash scripts/train.sh 0
bash scripts/eval.sh 0

Stitching Example

Note that stitching large-sized images may cause the GPU out-of-memory due to the consumption of the backbone.

bash scripts/stitch.sh left.jpg right.jpg

Acknowlegment

This work is mainly based on LTEW and IHN, we thank the authors for the contribution.