edge2view-demo

This is a pix2pix demo that learns from edge and translates this into view. A interactive application is also provided that translates edge to view.

Getting Started

1. Prepare Environment

# Clone this repo
git clone [email protected]:GordonRen/edge2view.git

# Create the conda environment from file
conda env create -f environment.yml

2. Configure Holistically-Nested Edge Detection

https://github.com/s9xie/hed

3. Generate Original Data

python generate_train_data.py --file Desert.mp4

Input:

• file is the name of the video file from which you want to create the data set.

Output:

• One folder original will be created.

4. Generate Edge Data

• generate edge data by following batch_hed.py and put the edge data into hed_edge.

If you want to download my dataset, here is also the video file that I used and the generated training dataset (708 images already split into training and validation).

5. Train Model

# Clone the repo from Christopher Hesse's pix2pix TensorFlow implementation
git clone https://github.com/affinelayer/pix2pix-tensorflow.git

# Move the original and hed_edge folder into the pix2pix-tensorflow folder
mv edge2view/hed_edge edge2view/original pix2pix-tensorflow/photos_view

# Go into the pix2pix-tensorflow folder
cd pix2pix-tensorflow/

# Reset to april version
git reset --hard d6f8e4ce00a1fd7a96a72ed17366bfcb207882c7

# Resize original images
python tools/process.py \
  --input_dir photos_view/original \
  --operation resize \
  --output_dir photos_view/original_resized
  
# Resize hed_edge images
python tools/process.py \
  --input_dir photos_view/hed_edge \
  --operation resize \
  --output_dir photos_view/hed_edge_resized
  
# Combine both resized original and hed_edge images
python tools/process.py \
  --input_dir photos_view/hed_edge_resized \
  --b_dir photos_view/original_resized \
  --operation combine \
  --output_dir photos_view/combined
  
# Split into train/val set
python tools/split.py \
  --dir photos_view/combined
  
# Train the model on the data
python pix2pix.py \
  --mode train \
  --output_dir edge2view-model \
  --max_epochs 1000 \
  --input_dir photos_view/combined/train \
  --which_direction AtoB

For more information around training, have a look at Christopher Hesse's pix2pix-tensorflow implementation.

6. Export Model

1. First, we need to reduce the trained model so that we can use an image tensor as input:

   python reduce_model.py --model-input edge2view-model --model-output edge2view-reduced-model
   

   Input:

   • model-input is the model folder to be imported.
   • model-output is the model (reduced) folder to be exported.

   Output:

   • It returns a reduced model with less weights file size than the original model.

2. Second, we freeze the reduced model to a single file.

   python freeze_model.py --model-folder edge2view-reduced-model
   

   Input:

   • model-folder is the model folder of the reduced model.

   Output:

   • It returns a frozen model file frozen_model.pb in the model folder.

I have uploaded a pre-trained frozen model here. This model is trained on 708 images with epoch 1000.

7. Run Demo

python edge2view.py --tf-model edge2view-reduced-model/frozen_model.pb

Input:

• tf-model is the frozen model file.

Example:

Requirements

• TensorFlow 1.0.0
• Holistically-Nested Edge Detection
• MATLAB

Acknowledgments

Kudos to Christopher Hesse for his amazing pix2pix TensorFlow implementation and Gene Kogan for his inspirational workshop.
Inspired by Dat Tran.

License

See LICENSE for details.