GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → AnatoliiPotapov → inpainting_FRRN

AnatoliiPotapov / inpainting_FRRN

Licence: other
Progressive Image Inpainting (Kolmogorov Team solution for Huawei Hackathon 2019 summer)

Programming Languages

Jupyter Notebook
11667 projects

Labels

torchpytorchganpix2pixinpaintingpatchgan

Projects that are alternatives of or similar to inpainting FRRN

tiny-pix2pix
Redesigning the Pix2Pix model for small datasets with fewer parameters and different PatchGAN architecture
Stars: ✭ 21 (-30%)
Mutual labels:  pix2pix, patchgan
Deepnudecli
DeepNude Command Line Version With Watermark Removed
Stars: ✭ 112 (+273.33%)
Mutual labels:  torch, pix2pix
Cyclegan
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
Stars: ✭ 10,933 (+36343.33%)
Mutual labels:  torch, pix2pix
Gpt2 Newstitle
Chinese NewsTitle Generation Project by GPT2.带有超级详细注释的中文GPT2新闻标题生成项目。
Stars: ✭ 235 (+683.33%)
Mutual labels:  torch
Pixeldtgan
A torch implementation of "Pixel-Level Domain Transfer"
Stars: ✭ 248 (+726.67%)
Mutual labels:  torch
pix2pix.pytorch
A pytorch implementation of "Image-to-Image Translation with Conditional Adversarial Networks"
Stars: ✭ 34 (+13.33%)
Mutual labels:  pix2pix
image-background-remove-tool
✂️ Automated high-quality background removal framework for an image using neural networks. ✂️
Stars: ✭ 767 (+2456.67%)
Mutual labels:  torch
Echotorch
A Python toolkit for Reservoir Computing and Echo State Network experimentation based on pyTorch. EchoTorch is the only Python module available to easily create Deep Reservoir Computing models.
Stars: ✭ 231 (+670%)
Mutual labels:  torch
chainer-pix2pix
Chainer implementation for Image-to-Image Translation Using Conditional Adversarial Networks
Stars: ✭ 40 (+33.33%)
Mutual labels:  pix2pix
ThArrays.jl
A Julia interface for PyTorch's C++ backend, focusing on Tensor, AD, and JIT
Stars: ✭ 23 (-23.33%)
Mutual labels:  torch
lantern
[Android Library] Handling device flash as torch for Android.
Stars: ✭ 81 (+170%)
Mutual labels:  torch
Artificial Intelligence Deep Learning Machine Learning Tutorials
A comprehensive list of Deep Learning / Artificial Intelligence and Machine Learning tutorials - rapidly expanding into areas of AI/Deep Learning / Machine Vision / NLP and industry specific areas such as Climate / Energy, Automotives, Retail, Pharma, Medicine, Healthcare, Policy, Ethics and more.
Stars: ✭ 2,966 (+9786.67%)
Mutual labels:  torch
deep-learning-for-document-dewarping
An application of high resolution GANs to dewarp images of perturbed documents
Stars: ✭ 100 (+233.33%)
Mutual labels:  pix2pix
Floorplantransformation
Raster-to-Vector: Revisiting Floorplan Transformation
Stars: ✭ 243 (+710%)
Mutual labels:  torch
MXNet-GAN
MXNet Implementation of DCGAN, Conditional GAN, pix2pix
Stars: ✭ 23 (-23.33%)
Mutual labels:  pix2pix
Alphaction
Spatio-Temporal Action Localization System
Stars: ✭ 221 (+636.67%)
Mutual labels:  torch
pytorch-gans
PyTorch implementation of GANs (Generative Adversarial Networks). DCGAN, Pix2Pix, CycleGAN, SRGAN
Stars: ✭ 21 (-30%)
Mutual labels:  pix2pix
graftr
graftr: an interactive shell to view and edit PyTorch checkpoints.
Stars: ✭ 89 (+196.67%)
Mutual labels:  torch
multiclass-semantic-segmentation
Experiments with UNET/FPN models and cityscapes/kitti datasets [Pytorch]
Stars: ✭ 96 (+220%)
Mutual labels:  torch
Everybody-dance-now
Implementation of paper everybody dance now for Deep learning course project
Stars: ✭ 22 (-26.67%)
Mutual labels:  pix2pix
View All Similar Projects ➔

Kolmogorov team

solution for Huawei inpainting hackathon https://huawei-hackathon.moscow/ !

UPD: Second place on public set and third place on private set. Leaderboard: http://bit.ly/huawei_hack_lb

Details

Core paper: Progressive Image Inpainting with Full-Resolution Residual Network https://arxiv.org/abs/1907.10478

Tricks and modifications:

Use Full Resolution Places 365 dataset (http://places2.csail.mit.edu) for training, with categories selected to be visually close on the hackathon train set.

Train two models (v2 and v2_discriminator) v2: with l1 loss and l1 step loss (for each block) v2_discriminator: v2 + PatchGAN + style losses

Final solution: blend of 11 models with equal weights and bluring (see predict.py)

Training

python train.py

Predict

python predict.py --pred_path ../Datasets/Huawei/DATASET_INPAINTING/new_result/ --config_path experiments/config_v2.yml --checkpoint experiments/checkpoints/ --masks_path experiments/masks/private/  --images_path ../Datasets/Huawei/DATASET_INPAINTING/test_final/ --cuda '1' --batch_size 10 --blured True

To reproduce our solution, you need to download solution.zip from https://drive.google.com/file/d/1BLVtkeiB2EF7mE61vALKV9yP3ti-akT_/view?usp=sharing, then unzip it to experiments/

PPS: for Dataset class to work each folder with pictures should contain files.txt with local filenames

Made by Tinkoff NLP team

Anatoly Potapov (*) (research and coordination, choice of paper, implementation of FRRB)

Vasily Karmazin (implemented most of the ideas, introduced bluring)

Gleb Ishelev (helped a lot)

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].