This is the official github repository for the paper "What's There in The Dark" accepted in IEEE International Conference in Image Processing 2019 (ICIP19) , Taipei, Taiwan. [ Paper][Supplementary] Papers with code ] [Poster]

Using this you can do semantic segmentation on night images.

Scene Parsing is an important cog for modern autonomous driving systems. Most of the works in semantic segmentation pertains to day-time scenes with favourable weather and illumination conditions. In this paper, we propose a novel deep architecture, NiSeNet, that performs semantic segmentation of night scenes using a domain mapping approach of synthetic to real data. It is a dual-channel network, where we designed a Real channel using DeepLabV3+ coupled with an MSE loss to preserve the spatial information. In addition, we used an Adaptive channel reducing the domain gap between synthetic and real night images, which also complements the failures of Real channel output. Apart from the dual channel, we introduced a novel fusion scheme to fuse the outputs of two channels. In addition to that, we compiled a new dataset Urban Night Driving Dataset (UNDD); it consists of 7125 unlabelled day and night images; additionally, it has 75 night images with pixel-level annotations having classes equivalent to Cityscapes dataset. We evaluated our approach on the Berkley Deep Drive dataset, the challenging Mapillary dataset and UNDD dataset to exhibit that the proposed method outperforms the state-of-the-art techniques in terms of accuracy and visual quality

Since we are submitting in journal, we currently cannot make the code public, however, we are making the data preparation code public. Infact we are the first one to make the model for day to night image conversion public. However, we have made the multi-scale architecture code public. Interested users can download adaptsegnet and deeplabv3+ and plug this code in as the last module and train.

• Go to the CycleGANS github link and clone the code.
• Download the pretrained model from the following link [ Pretrained_Model ].
• Place the folder contents (latest_net_G_A.pth, latest_net_G_B.pth, latest_net_D_A.pth, latest_net_D_B.pth) into the checkpoint/any_name folder.
• Run the testing code as mentioned in CycleGANS website.

• Step 1 : Cloning and Environment Setup :

    python3 -m venv icip
    source activate icip
    git clone https://github.com/sauradip/night_image_semantic_segmentation.git
    cd night_image_semantic_segmentation
    pip3 install -r icip_requirements.txt

• Step 2 : Downloading weights for Real Channel and Adaptive Channel

    Place the "checkpoints" dir from Real Channel Link in "real/DeepLabV3Plus-Pytorch"
    Place the files in Adaptive CHannel Link in "adapt/AdaptSegNet/model"

• Step 3 : Prepare Data ( A sample has been given in "data_" directory )

    Place the preprocessed data mentioned in preprocessing step in "data_" folder , i.e preprocessed Night Images in "leftImg8bit" folder and GT In "gtFine" folder

• Step 4 : Set Paths and other hyperparameters in file "config/cityscapes_config.py"

IMPORTANT

If you want to run this code for other datasets, train the real channel and adaptive channel with their respective training codes given in "real/DeepLabV3Plus-Pytorch" and "adapt/AdaptSegNet" folder and follow Step 2 and Step 3. I have provided checkpoints for Cityscapes only

• Step 5 : Training

    python main_model.py 

The checkpoints will be stored in "real_checkpoint" directory ( Since we will use only real channel weights during inference

• Step 6 : Testing

   sh test_v2.sh

Some sample results and data format samples have been provided in this link

The following link contains our proposed Urban Night Driving Dataset. We have provided only the testing split for now

The results were generated using NiSeNet on the following dataset BDD, Mapillary, UNDD(proposed) :

Updated codes and models are uploaded in the following link Code Will add in the main repository once free.

If you find this code useful in your research, please consider citing:

    @inproceedings{nag2019s,
    title={What’s There in the Dark},
    author={Nag, Sauradip and Adak, Saptakatha and Das, Sukhendu},
    booktitle={2019 IEEE International Conference on Image Processing (ICIP)},
    pages={2996--3000},
    year={2019},
    organization={IEEE}
  }