cuhk-eda/neural-ilt

Neural-ILT, An End-to-end Learning-based Mask Optimizer

neural-ilt's Issues

Inquiry Regarding Image Scaling and Cropping Methodology

Hello Neural-ILT team,

I've been delving into the Neural-ILT project and am quite impressed with the work you've achieved. While examining the image processing steps taken prior to inputting images into the network, I noted a distinct approach to scaling and cropping. This approach piqued my interest, as it appears to diverge from traditional methods that typically involve resizing images to a uniform dimension to preserve scale, followed by a process of random cropping.

Could you please provide some insight into the rationale behind this preprocessing strategy? Specifically, I'm curious about the reasons for saving different values for each image and applying custom cropping. Is this approach essential? Additionally, considering there's scaling involved, could you inform me about the image scale at which the kernels are designed to operate, especially if I'm considering doing custom cropping?

Thank you for your time and consideration. I eagerly await your response :)

Thomas,

When will the code of Neural Ilt 2.0 be open-sourced

Hi @liulixinkerry , thank you for your work, I see the paper "Neural Ilt 2.0" is released that add L_epe_fast loss as DSR term , how is EPE computing implemented? and when will the code of Neural Ilt 2.0 be open-sourced？

Runtime Error

Hi Authors, thanks for the work. I am trying to run the example using python .\neural_ilt.py, but I got the following error information:

RuntimeError: Cannot initialize CUDA without ATen_cuda library. PyTorch splits its backend into two shared libraries: a CPU library and a CUDA library; this error has occurred because you are trying to use some CUDA functionality, but the CUDA library has not been loaded by the dynamic linker for some reason. The CUDA library MUST be loaded, EVEN IF you don't directly use any symbols from the CUDA library! One common culprit is a lack of -INCLUDE:?warp_size@cuda@at@@yahxz in your link arguments; many dynamic linkers will delete dynamic library dependencies if you don't depend on any of their symbols. You can check if this has occurred by using link on your binary to see if there is a dependency on *_cuda.dll library.

My environment (Win64, anaconda, 3090-Ti) is as follows:

blas                      1.0                         mkl
ca-certificates           2022.12.7            h5b45459_0    conda-forge
certifi                   2022.12.7          pyhd8ed1ab_0    conda-forge
cudatoolkit               11.1.1              heb2d755_10    conda-forge
eigen                     3.3.7                h59b6b97_1
glib                      2.69.1               h5dc1a3c_2
gst-plugins-base          1.18.5               h9e645db_0
gstreamer                 1.18.5               hd78058f_0
hdf5                      1.12.1               h1756f20_2
icc_rt                    2022.1.0             h6049295_2
icu                       58.2                 ha925a31_3
intel-openmp              2021.4.0          haa95532_3556
jpeg                      9e                   h2bbff1b_0
lerc                      3.0                  hd77b12b_0
libclang                  12.0.0          default_h627e005_2
libdeflate                1.8                  h2bbff1b_5
libffi                    3.4.2                hd77b12b_6
libiconv                  1.16                 h2bbff1b_2
libogg                    1.3.5                h2bbff1b_1
libpng                    1.6.37               h2a8f88b_0
libprotobuf               3.20.1               h23ce68f_0
libtiff                   4.4.0                h8a3f274_2
libvorbis                 1.3.7                he774522_0
libwebp                   1.2.4                h2bbff1b_0
libwebp-base              1.2.4                h2bbff1b_0
libxml2                   2.9.14               h0ad7f3c_0
libxslt                   1.1.35               h2bbff1b_0
lz4-c                     1.9.4                h2bbff1b_0
mkl                       2021.4.0           haa95532_640
mkl-service               2.4.0            py37h2bbff1b_0
mkl_fft                   1.3.1            py37h277e83a_0
mkl_random                1.2.2            py37hf11a4ad_0
numpy                     1.21.5           py37h7a0a035_3
numpy-base                1.21.5           py37hca35cd5_3
opencv                    4.6.0            py37h104de81_2
opencv-python             4.7.0.68                 pypi_0    pypi
openssl                   1.1.1s               h2bbff1b_0
pcre                      8.45                 hd77b12b_0
pillow                    6.1.0                    pypi_0    pypi
pip                       22.3.1           py37haa95532_0
python                    3.7.3                h8c8aaf0_1
qt-main                   5.15.2               he8e5bd7_7
qt-webengine              5.15.9               hb9a9bb5_4
qtwebkit                  5.212                h3ad3cdb_4
setuptools                65.5.0           py37haa95532_0
six                       1.16.0             pyhd3eb1b0_1
sqlite                    3.40.0               h2bbff1b_0
torch                     1.8.0                    pypi_0    pypi
torchvision               0.2.2                    pypi_0    pypi
tqdm                      4.19.9                   pypi_0    pypi
typing-extensions         4.4.0                    pypi_0    pypi
vc                        14.2                 h21ff451_1
vs2015_runtime            14.27.29016          h5e58377_2
wheel                     0.37.1             pyhd3eb1b0_0
wincertstore              0.2              py37haa95532_2
xz                        5.2.8                h8cc25b3_0
zlib                      1.2.13               h8cc25b3_0
zstd                      1.5.2                h19a0ad4_0

Is there any reason for the error? Thanks!

Why gamma=4?

Dear authors,

Hi, thanks for the work. I noticed that when you compute the ILT loss, the value of gamma you used is 4, instead of commonly used value of 2:
ilt_loss = (result - target).pow(4).sum()

Is there a specific reason that this value is used?

Thanks!

origin of the kernels

The .pt files seems to be the kernels related to the lithography system, is there any code to generate them instead of simply loading?