SegFormer Open-Source Lane Detection Model - Lightweight Design, Free and Supports Precise Lane Recognition

Segformer Finetuned Lane 10k Steps

Developed by Efferbach

A lightweight lane line detection model based on the SegFormer architecture, fine-tuned for 10,000 steps on the lane_master dataset

Image Segmentation

Transformers

Open Source License:Other #Lane line segmentation #Traffic scene recognition #Lightweight model

Downloads 1,077

Release Time : 4/7/2023

Model Overview

This model is a fine-tuned version of NVIDIA SegFormer-b0, pre-trained on the Cityscapes dataset, specifically optimized for lane line detection tasks. It is designed to identify left and right lane markings in road scenes.

Model Features

Efficient lane line detection

Optimized for road scenes, accurately identifying left and right lane lines

Lightweight architecture

Designed based on SegFormer-b0's lightweight structure, suitable for real-time applications

Transfer learning optimization

Fine-tuned on a Cityscapes pre-trained model to enhance lane line detection performance

Model Capabilities

Road scene image segmentation

Left and right lane line recognition

Pixel-level semantic segmentation

Use Cases

Autonomous driving

Lane keeping assist system

Real-time detection of vehicle lane position

Average IoU 0.49, accuracy 0.737

Road scene analysis

Extracting structured road information for high-definition map construction

🚀 segformer-finetuned-lane-10k-steps

This model is a fine - tuned version of nvidia/segformer-b0-finetuned-cityscapes-512-1024 on the Efferbach/lane_master dataset, used for image segmentation tasks.

🚀 Quick Start

This model is a fine-tuned version of nvidia/segformer-b0-finetuned-cityscapes-512-1024 on the Efferbach/lane_master dataset. It achieves the following results on the evaluation set:

Loss: 0.0365
Mean Iou: 0.4899
Mean Accuracy: 0.7371
Overall Accuracy: 0.7371
Accuracy Background: nan
Accuracy Left: 0.7394
Accuracy Right: 0.7348
Iou Background: 0.0
Iou Left: 0.7371
Iou Right: 0.7325

📚 Documentation

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 6e-05
train_batch_size: 8
eval_batch_size: 8
seed: 1337
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: polynomial
training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Background	Accuracy Left	Accuracy Right	Iou Left	Iou Right
0.0792	1.0	308	0.0714	0.0148	0.0229	0.0225	nan	0.0373	0.0085	0.0362	0.0083
0.0437	2.0	616	0.0502	0.1687	0.2775	0.2784	nan	0.2492	0.3058	0.2343	0.2718
0.0326	3.0	924	0.0445	0.2614	0.4441	0.4479	nan	0.3134	0.5748	0.3100	0.4742
0.0224	4.0	1232	0.0370	0.4048	0.6098	0.6100	nan	0.6043	0.6153	0.6031	0.6113
0.0184	5.0	1540	0.0346	0.3820	0.5858	0.5870	nan	0.5421	0.6295	0.5400	0.6060
0.0159	6.0	1848	0.0319	0.4367	0.6567	0.6573	nan	0.6343	0.6791	0.6341	0.6760
0.0139	7.0	2156	0.0317	0.4555	0.6855	0.6860	nan	0.6691	0.7019	0.6680	0.6986
0.0129	8.0	2464	0.0321	0.4348	0.6533	0.6535	nan	0.6479	0.6588	0.6474	0.6571
0.0122	9.0	2772	0.0275	0.4541	0.6827	0.6830	nan	0.6710	0.6943	0.6697	0.6927
0.0111	10.0	3080	0.0305	0.4609	0.6928	0.6927	nan	0.6969	0.6887	0.6963	0.6865
0.011	11.0	3388	0.0286	0.4646	0.6988	0.6991	nan	0.6890	0.7087	0.6883	0.7055
0.0103	12.0	3696	0.0298	0.4693	0.7058	0.7062	nan	0.6939	0.7177	0.6932	0.7148
0.0097	13.0	4004	0.0293	0.4717	0.7090	0.7087	nan	0.7184	0.6996	0.7176	0.6975
0.0093	14.0	4312	0.0330	0.4537	0.6835	0.6836	nan	0.6775	0.6894	0.6768	0.6843
0.009	15.0	4620	0.0331	0.4804	0.7226	0.7226	nan	0.7194	0.7257	0.7178	0.7234
0.0088	16.0	4928	0.0315	0.4890	0.7355	0.7357	nan	0.7275	0.7435	0.7259	0.7411
0.0086	17.0	5236	0.0338	0.4813	0.7234	0.7234	nan	0.7224	0.7243	0.7216	0.7223
0.0085	18.0	5544	0.0348	0.4743	0.7129	0.7126	nan	0.7225	0.7033	0.7217	0.7012
0.0083	19.0	5852	0.0357	0.4812	0.7245	0.7244	nan	0.7281	0.7210	0.7254	0.7183
0.0081	20.0	6160	0.0334	0.4829	0.7271	0.7269	nan	0.7337	0.7205	0.7305	0.7182
0.0079	21.0	6468	0.0359	0.4773	0.7177	0.7177	nan	0.7184	0.7170	0.7174	0.7146
0.0077	22.0	6776	0.0351	0.4874	0.7332	0.7329	nan	0.7440	0.7223	0.7432	0.7190
0.0075	23.0	7084	0.0344	0.4855	0.7296	0.7292	nan	0.7437	0.7156	0.7425	0.7141
0.0077	24.0	7392	0.0362	0.4799	0.7216	0.7216	nan	0.7236	0.7196	0.7223	0.7174
0.0071	25.0	7700	0.0391	0.4775	0.7179	0.7180	nan	0.7173	0.7186	0.7161	0.7163
0.0077	26.0	8008	0.0339	0.4895	0.7367	0.7366	nan	0.7405	0.7329	0.7388	0.7297
0.0069	27.0	8316	0.0344	0.4858	0.7305	0.7305	nan	0.7291	0.7318	0.7278	0.7297
0.0069	28.0	8624	0.0361	0.4844	0.7283	0.7282	nan	0.7324	0.7243	0.7309	0.7221
0.007	29.0	8932	0.0371	0.4837	0.7273	0.7270	nan	0.7360	0.7186	0.7345	0.7166
0.007	30.0	9240	0.0366	0.4854	0.7305	0.7303	nan	0.7379	0.7231	0.7353	0.7208
0.0067	31.0	9548	0.0367	0.4866	0.7322	0.7321	nan	0.7357	0.7286	0.7335	0.7263
0.0068	32.0	9856	0.0364	0.4883	0.7348	0.7347	nan	0.7377	0.7318	0.7355	0.7295
0.0067	32.47	10000	0.0365	0.4899	0.7371	0.7371	nan	0.7394	0.7348	0.7371	0.7325

Framework versions

Transformers 4.28.0.dev0
Pytorch 2.0.0+cu118
Datasets 2.11.0
Tokenizers 0.13.3

📄 License

other

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご