cnn_depth_tensorflow

cnn_depth_tensorflow is an implementation of depth estimation using tensorflow.

Presentation slides of this project: https://docs.google.com/presentation/d/1ytuDdWJUG9VTK7wCVm3BzbW1uuGI_sRgH0TCZc1PNe0/edit?usp=sharing

Reference

Original paper is "Depth Map Prediction from a Single Image using a Multi-Scale Deep Network". https://arxiv.org/abs/1406.2283

Modified from: https://github.com/MasazI/cnn_depth_tensorflow

Requierments

• TensorFlow 1.3 with python 2.7
• Numpy
• Pillow

File descriptions

• Depth.py defines the model and training operations. We built our computational graph and other training details here.
• Test.py is for testing using any .jpg input.

Testing

1. Put your .jpg image at 'data/'
2. Make sure your filename appears in test_test.csv
3. Run

python2 Test.py

4. To output coarse results, edit Test.py:

REFINE_TRAIN = True

Training

Since the original dataset is not included here, before training, you may have to first download http://horatio.cs.nyu.edu/mit/silberman/nyu_depth_v2/nyu_depth_v2_labeled.mat and drop it at 'data/'. Then run

python convert_mat_to_img.py

Then run

python2 Depth.py

Some configuration:

Coarse train: REFINE_TRAIN = False FINE_TUNE = False

Fine train: REFINE_TRAIN = True FINE_TUNE = True

Coarse train should come first.