- Approximately 45,500 images, 2.3GB. One of the original datasets made by SullyChen in 2017. Data was recorded around Rancho Palos Verdes and San Pedro California.
- 25 minutes = 25{min} x 60{1 min = 60 sec} x 30{fps} = 45,000 images ~ 2.3 GB
- You Can Download Dataset from Here
- How the data was recorded by SullyChen can is explained in one his medium articles. link to SullyChen medium article
- You can see how the images are recorded from this Video.
- self_driving_using_keras.ipynb file contains 4 different Architectures trained on the data for 100 epochs and selected the 2 best models that converge loss faster.
- Best_Model_1.ipynb , Best_Model_2.ipynb contains the model trained for 1000 epochs.
- Best_Model_1 and Best_Model_2 files contains CSV,Excel,Json files of history object showing loss on epoch 1-1000. The Output Model is also saved in it. -run_model_one and run_model_two are Python files for running the model on dataset.
Got Best Results Using Nvidia's Architecture. It is used in Best_Model_1
These are the Results after 1000 epochs.
-
val_loss loss min 0.155158 0.006902 max 0.314412 0.264360 mean 0.175964 0.011505
| val_loss | loss | |
|---|---|---|
| min | 0.170120 | 0.009587 |
| max | 0.822659 | 0.292208 |
| mean | 0.316213 | 0.014819 |
- Udacity:
70 minutes of data ~ 223GB
Format: Image, latitude, longitude, gear, brake, throttle, steering angles and speed
- Udacity Dataset: Dataset ranging from 40 to 183 GB in different conditions
- Comma.ai Dataset [80 GB Uncompressed]
- Apollo Dataset with different environment data of road
-
SullyChen github page
-
cyanamous github page
-
Research paper: End to End Learning for Self-Driving Cars by Nvidia.
-
Nvidia blogs:
-
https://devblogs.nvidia.com/deep-learning-self-driving-cars/
-
https://devblogs.nvidia.com/explaining-deep-learning-self-driving-car/