Skip to content

Neural Radiance Field (ECCV2020), TF2 & Keras implementation

Notifications You must be signed in to change notification settings

ProtossDragoon/NeRF-TF2-Keras

Repository files navigation

NeRF Tensorflow2 & Keras Implementation

Neural Radience Field Tensorflow2 & Keras implementation

Quickstart

1. Download dataset

./download_example_data.bash

2. Install dependencies

virtual environment

python3 -m venv ~/dev/tf2
source ~/dev/tf2/bin/activate

requirements

pip install -r requirements.txt

3. Run unit tests

tests

python3 sampledata_loader_test.py
python3 train_test.py
python3 -m nerf_core.ray_test
python3 -m nerf_utils.videovis_test
  • If there aren't any error, go to the next step.
  • When you found a bug, please raise an github issue.

4. Run demo

python3 train.py

Results

Monitor Trainig
40epochs
32samples/ray
20epoch_32samples
80epochs
64samples/ray
80epoch_64samples
Visualization pitch=-30.0 pitch=-10.0 pitch=+10.0
40epochs
32samples/ray
16posenc
5batch
40epoch_32samples 40epoch_32samples 40epoch_32samples
80epochs
64samples/ray
16posenc
5batch
80epoch_64samples 80epoch_64samples 80epoch_64samples
160epochs
64samples/ray
32posenc
1batch
160epoch_64samples_32positionalencoding_1batch 160epoch_64samples_32positionalencoding_1batch 160epoch_64samples_32positionalencoding_1batch

Note

1. File Structure

  • FILLME

2. Implementataion

  • Hierarchical View Sampling
  • Positional Encoding
  • Transform Ray to NDC

3. Test Environments

  • 19' MacBook Pro 16inch, 16GB
  • Google COLAB GPU Runtime

References