CIS6930-Project-2

• This Project tries to do deep colorization to transfer from the grayscale image into RGB scale image.

Environmental setting recommendation

• Recommend to use Anaconda 64-bit to build the environment for applying CUDA to run GPU and used the Jupyter notebook from Anaconda to build Pytorch programs.

• Recommend to use Anaconda with the Jupyter notebook to run our programs with the same environmental setting.

• Below are folders containing each file are expressed (Folder-> Each file).

Programs

• The main programs are the RegressorColorizing.ipynb and ColorizingSigmoidChangeFM.ipynb for the project.

1. face_images: a dataset of 750 images.
2. ColorizingRelu.ipynb: Input L*, and output matrix a* and b* with ReLU last layer. Then, colorizing the images.
3. ColorizingSigmoid.ipynb: Input L*, and output matrix a* and b* with Sigmoid last layer. Then, colorizing the images.
4. ColorizingSigmoidChangeFM.ipynb: Input L*, output matrix a* and b* with Sigmoid last layer, and change each layers' feature maps. Then, colorizing the images.
5. ColorizingTanh.ipynb: Input L*, output matrix a* and b* with Tanh last layer. Then, colorizing the images.
6. RegressorColorizing.ipynb: Input L* and predicted mean scalar a* and b* by regressor. Output matrix a* and b* with Sigmoid last layer. (3 input factors L*, scalar a*, and b*; 2 output factors matrix a* and b*)
7. Regressor.ipynb: Input L*, prediction mean scalar a* and b*.
8. ImplementTrainedModel4Colorizing.ipynb: Implement trained parameters to color images without training again. Only for ColorizingSigmoid, ColorizingSigmoidChangeFM, and ColorizingRelu trained model. Please select a correct net model for corresponded trained parameters.
9. ImplementTrainedModel4CombRegClo.ipynb: Implement trained parameters to color images without training again. Only for RegressorColorizing trained model.

TrainedModel

• These archives are trained models and can be applied by ImplementTrainedModel4Colorizing.ipynb and ImplementTrainedModel4CombRegClo.ipynb to redrawing images.

1. ColorizingReludWithLR0.1EP1000.pth: Trained model of ColorizingRelu.ipynb program.
2. ColorizingSigmoidWithLR0.1EP1000.pth: Trained model of ColorizingSigmoid.ipynb program.
3. ColorizingSigmoidChangeFMTrainTestLogEP1000.pth: Trained model of ColorizingSigmoidChangeFM.ipynb program.
4. ColorizingTanhdWithLR0.1EP1000.pth: Trained model of ColorizingTanh.ipynb program.
5. ColorizingCombWithLR0.1EP1000: Trained model of RegressorColorizing.ipynb program for colorizing model.
6. RegCombWithLR0.1EP1000.pth: Trained model of RegressorColorizing.ipynb program for regression model.
7. RegWithLR0.1EP1000.pth: Trained model of Regressor.ipynb program.

TrainTestImagesRes

• Test results are the same images.

1. ColorizingRelu--.png: Training and testing image results of ColorizingRelu.ipynb.
2. ColorizingSigmoid--.png: Training and testing image results of ColorizingSigmoid.ipynb.
3. ColorizingSigmoidChangeFM--.png: Training and testing image results of ColorizingSigmoidChangeFM.ipynb.
4. ColorizingTanh--.png: Training and testing image results of ColorizingTanh.ipynb.
5. ColorizingComb--.png: Training and testing image results of RegressorColorizing.ipynb for colorizing model.

TrainTestLog

• First to last the second line are training records, and the last line is testing records.

1. ColorizingReluTrainTestLog1000.txt: The training and testing log of ColorizingRelu.ipynb.
2. ColorizingSigmoidTrainTestLog1000.txt: The training and testing log of ColorizingSigmoid.ipynb.
3. ColorizingSigmoidChangeFMTrainTestLogEP1000.txt: The training and testing log of ColorizingSigmoidChangeFM.ipynb.
4. ColorizingTanhTrainTestLogEP1000.txt: The training and testing log of ColorizingTanh.ipynb.
5. ColorizingCombTrainTestLogEP1000.txt: The training and testing log of RegressorColorizing.ipynb program for colorizing model.
6. RegCombTrainTestLogEP1000.txt: The training and testing log of RegressorColorizing.ipynb program for the regression model.
7. RegTrainTestLog1000.txt: The training and testing log of Regressor.ipynb program.