文件列表:

evaluation.ipynb (54513, 2023-10-14)
example_plots.ipynb (638629, 2023-10-14)
multi_classes_plots.ipynb (801692, 2023-10-14)
u2_net.py (15241, 2023-10-14)
u2_net_train.py (14161, 2023-10-14)
unet_segmentation.py (16250, 2023-10-14)
unet_segmentation_multi.py (19300, 2023-10-14)

# image_segmentation # Image Segmentation using a U-Net * forest_segmentation: binary segmentation * **data: https://www.kaggle.com/datasets/quadeer15sh/augmented-forest-segmentation** * model: unet_segmentation.py * input is expected to be in folders 'masks' and 'images' as .jpg images * cityscapes-images-paris: multiple segmentation * data: https://www.kaggle.com/datasets/dansbecker/cityscapes-image-pairs * preprocessing of the multiclass masks has been adapted from: https://www.kaggle.com/code/yauhenikavaliou/camseq-semantic-segmentation * model: unet_segmentation_multi.py * input is expected to be in folders 'masks' and images * Possible improvements: * add class weights --- # Semantic Image Segmentation using U-Net and CNNs for Self-Driving Cars ## Overview This repository contains the implementation of a U-Net architecture for semantic image segmentation on the CARLA self-driving car dataset using deep learning and Convolutional Neural Networks (CNNs). Semantic image segmentation is a critical task in autonomous driving, enabling vehicles to understand their surroundings by classifying each pixel in an image into predefined categories. ## Technologies Used - **Framework:** TensorFlow - **Architecture:** U-Net - **Accuracy:** 97% ## Key Features - **U-Net Architecture:** Implemented a U-Net model from scratch, a widely used architecture in image segmentation tasks due to its effectiveness in capturing fine-grained details. - **CARLA Dataset:** Utilized the CARLA self-driving car dataset, a diverse and challenging dataset for training and evaluating the segmentation model. - **Deep Learning:** Leveraged Convolutional Neural Networks (CNNs) to learn spatial hierarchies and features for accurate pixel-wise predictions. - **High Accuracy:** Achieved an impressive accuracy of 97% in semantic image segmentation, demonstrating the model's ability to accurately classify objects in the environment. ## How to Use 1. **Dataset Preparation:** Download the CARLA self-driving car dataset and preprocess the images and corresponding segmentation masks for training. 2. **Training:** Train the U-Net model using TensorFlow, adjusting hyperparameters as necessary for optimal performance. 3. **Evaluation:** Evaluate the trained model on test data to assess its segmentation accuracy and make any necessary improvements. 4. **Inference:** Use the trained model for real-time or batch inference on new images to perform semantic image segmentation. ## Results The trained U-Net model exhibited remarkable performance with an accuracy of 97% in classifying objects in the CARLA dataset. The segmentation results demonstrate the model's ability to accurately delineate objects such as vehicles, pedestrians, and road markings, showcasing its potential for real-world applications in self-driving cars. ## Future Work - **Fine-Tuning:** Explore fine-tuning strategies to further enhance the model's accuracy, especially in challenging scenarios such as adverse weather conditions and complex urban environments. - **Real-Time Inference:** Optimize the model for real-time inference, ensuring low latency and high throughput to enable seamless integration into autonomous vehicles. - **Data Augmentation:** Implement advanced data augmentation techniques to augment the training dataset, enhancing the model's ability to generalize to diverse driving scenarios. ---

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：