Android_Demo

Taichi NGP Android Demo

Our Android Demo is a semi-finished product at this moment, due to lack of expertise in Android development.

We support the entire infrastructure from setting up the camera position to finally generate the inferenced image. However the camera position is directly set in the code, while the inferenced image is written to the disk. To make it a real Demo, consider connecting the camera position with a Touch Event, and pass the inferenced image to Android GUI.

Please be aware that there are known issues with certain Android devices, which is primarily related to their Vulkan support.

There's also performance optimization opportunities wrt Android deployment.

We welcome all contributions to improve the quality of the Android Demo!

Prerequisites

1. Device running Android with Vulkan support
2. Android Studio: https://developer.android.com/studio
3. Android Studio - adb: https://developer.android.com/tools/adb
4. Export ANDROID_NDK_ROOT and ANDROID_SDK_ROOT, for example:
5. Install Ninja: sudo apt-get update && sudo apt-get install ninja-build

export ANDROID_NDK_ROOT=~/Android/Sdk/ndk/25.1.8937393/
export ANDROID_SDK_ROOT=~/Android/Sdk/

Build and Install the Demo

1. Change camera position if neccessary open main.cpp and modify the following lines:

float angle_x, angle_y, angle_z = 0.0;
float radius = 2.5;
app.pose_rotate_scale(angle_x, angle_y, angle_z, radius);

2. Build and run the inference

sh scripts/compile_and_run_nerf.sh

3. Checkout the output image

• You'll notice a out.png generated in the same folder, which contains the inferenced image with current camera position.

Train and deploy your own NGP model

Follow instructions from: How to train a deployable NGP model

In the end, you will obtain a bunch of AOT files at taichi_ngp/compiled

4. Modify the width and height in C++ code if you've changed --res_w or res_h:

• open main.cpp and modify the following lines:

int img_width  = CUSTOM_WIDTH;
int img_height = CUSTOM_HEIGHT;
app.initialize(img_width, img_height,
               aot_file_root,
               hash_embedding_path,
               sigma_weights_path,
               rgb_weights_path,
               density_bitfield_path,
               pose_path,
               directions_path);

5. Install the Demo following steps from Build and Install the Demo and have fun!