The power predictor is part of the program I wrote for 'vibration-based harvester simulator' as author's final year project (FYP) for predicting the power generated if the car has the vision of its front view (using Unity3D built-in camera)
An updating version of power predictor is shown in this project.
- The heightmap (W:512 x H:64) regarding each point in an image in Unity unit is collected.
- The preprocessing module for heightmap (heightmap is recorded in a string format and the module is responsible for transfering the string into appropriate numpy format)
- The normalization module for heightmap (to normalize each pixel value into the range of [-1, 1]
- The model for predicting the heightmap from image has been updated to UNet-like structure
Dataset/ImageCollection: the image collected in UnityDataset/PowerCollection: average power collected and the heightmap stored in csvModel/: store the structure of the predictor modelPreprecessing/: store the utility for preprocessing, including:- extracting matrix data from csv file
- normalize matrix data
- power classification
In main.py, uncomment each block of code to implement different task. In github, the power classification task has been finished, meaning that the STEP3 is not necessary.
from Preprocessing import *
from Model import *
if __name__ == '__main__':
"""
STEP1: preprocessing the matrix recorded in the csv file
store the matrix in a specific location
"""
# LP = LabelPreprocessing.LabelPreprocessing(
# data_file_path='Dataset/PowerCollection',
# image_size=(512, 64)
# )
# LP.start()
"""
STEP2: normalize the value
"""
# norm = Normalization.HeightNormalization(
# data_file_path='Dataset/PowerCollection',
# custom_path=False
# )
# norm.start()
"""
STEP3: classify the power group
redistribute the images to specific location
"""
# pp = PowergroupPreprocessing.PowergroupPreprocessing(
# power_data_csv_path='Dataset/PowerCollection',
# src_folder_path='Dataset/ImageCollection',
# train_dst_folder_path='Train',
# test_dst_folder_path='Test',
# train_size=10000,
# num_classes=5,
# after_processing_data_file='Dataset/data_after_preprocessed.csv',
# multi_csv=True
# )
# pp.start()
"""
STEP4: train the model using
the image collected and
its correponding heightmap
"""
predictor = Predictor.EnergyPredictorV2_image2heightmap(
train_src_folder_path='Train',
test_src_folder_path='Test',
heightmap_src_folder_path='Dataset/PowerCollection/Label_heightmap_normalized',
checkpoint_folder_path='Checkpoint'
)
predictor.start()
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent