Autopilot
Autopilot copied to clipboard
akshaybahadur21
Steering not turning for the Autopilotv2 and i am using the same run video and same code
Yes ita working now.
On Tue, Nov 5, 2019, 3:37 PM maxmstrmn [email protected] wrote:
Did you succeed in the meantime?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBC5WRMQXEH7UAMVXJ3QSEPEBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDB4EXQ#issuecomment-549700190, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBAB7JVF3CCWCAOQPNLQSEPEBANCNFSM4HI6765Q .
Hi,
I didn't change anything. We just need to delete the label and features file from the folder and run it again. Load the file again, train it and run the AutopilotV2 file.
On Fri, Nov 8, 2019, 9:53 PM maxmstrmn [email protected] wrote:
What did you do? Could you share the source code?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBFB6LE754LBXTDDEJTQSVVMDA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDSDZZI#issuecomment-551828709, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBHCNTTCRCR3KFMYMALQSVVMDANCNFSM4HI6765Q .
Hi, thank you for this hint. But it doesn't work. I only use the v2 files from this repo (which don't include labels), train and test. The final file size of the h5 file is around 8mb, which is wrong as the file shipped in this repo ist 28mb large. Also if I print the network summary of both files it seems that some layers are missing in my file... Which environment do you use?
HI,
Yes it does not have labels and features. You must run load data.py first. Then it will create labels and features. For that, you must download a dataset from sully chen and store it inside the Autopilotv2 folder. Then run train.py. The model will be trained and stored in .h file. After that, run the AutopilotAppv2.py. It will work
On Thu, 21 Nov 2019 at 17:56, maxmstrmn [email protected] wrote:
Hi, thank you for this hint. But it doesn't work. I only use the v2 files from this repo (which don't include labels), train and test. The final file size of the h5 file is around 8mb, which is wrong as the file shipped in this repo ist 28mb large. Also if I print the network summary of both files it seems that some layers are missing in my file... Which environment do you use?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBDVJ3FJPCVCJFPLQL3QUZLLBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEEZUJ6Q#issuecomment-557008122, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBCLPQOJ7BOQG5NVQLTQUZLLBANCNFSM4HI6765Q .
Hi, I ran Load_DataV2.py on the dataset with 45000 images. This creates the labels and features files (~2GB size). Then I run the training script which creates the .h5 file, but this file is not as large as the file in this repo. I keep trying using the larger dataset, but there must be something wrong. I use Windows 10, Python 3.6, Keras 3.2.1 Tf 2.0.0 GPU.
No success, even with the larger dataset the final .h5 file is around 8mb large.
It seems the training is crazy. I printed both model summaries using the print_summary method from the keras utils. As you can see, in my trained model, a lot of layers are missing!
Here ist the output:
summary .h5 from this repo Model: "sequential_1"
Layer (type) Output Shape Param #
lambda_1 (Lambda) (None, 100, 100, 1) 0
conv2d_1 (Conv2D) (None, 100, 100, 32) 320
activation_1 (Activation) (None, 100, 100, 32) 0
max_pooling2d_1 (MaxPooling2 (None, 50, 50, 32) 0
conv2d_2 (Conv2D) (None, 50, 50, 64) 18496
activation_2 (Activation) (None, 50, 50, 64) 0
max_pooling2d_2 (MaxPooling2 (None, 25, 25, 64) 0
conv2d_3 (Conv2D) (None, 25, 25, 128) 73856
activation_3 (Activation) (None, 25, 25, 128) 0
max_pooling2d_3 (MaxPooling2 (None, 12, 12, 128) 0
flatten_1 (Flatten) (None, 18432) 0
dropout_1 (Dropout) (None, 18432) 0
dense_1 (Dense) (None, 128) 2359424
dense_2 (Dense) (None, 64) 8256
dense_3 (Dense) (None, 1) 65
Total params: 2,460,417 Trainable params: 2,460,417 Non-trainable params: 0
summary .h5 from my training output
Model: "sequential_1"
Layer (type) Output Shape Param #
lambda_1 (Lambda) (None, 100, 100, 1) 0
conv2d_1 (Conv2D) (None, 100, 100, 32) 320
activation_1 (Activation) (None, 100, 100, 32) 0
max_pooling2d_1 (MaxPooling2 (None, 50, 50, 32) 0
conv2d_2 (Conv2D) (None, 50, 50, 32) 9248
activation_2 (Activation) (None, 50, 50, 32) 0
max_pooling2d_2 (MaxPooling2 (None, 25, 25, 32) 0
conv2d_3 (Conv2D) (None, 25, 25, 64) 18496
activation_3 (Activation) (None, 25, 25, 64) 0
max_pooling2d_3 (MaxPooling2 (None, 12, 12, 64) 0
conv2d_4 (Conv2D) (None, 12, 12, 64) 36928
activation_4 (Activation) (None, 12, 12, 64) 0
max_pooling2d_4 (MaxPooling2 (None, 6, 6, 64) 0
conv2d_5 (Conv2D) (None, 6, 6, 128) 73856
activation_5 (Activation) (None, 6, 6, 128) 0
max_pooling2d_5 (MaxPooling2 (None, 3, 3, 128) 0
conv2d_6 (Conv2D) (None, 3, 3, 128) 147584
activation_6 (Activation) (None, 3, 3, 128) 0
max_pooling2d_6 (MaxPooling2 (None, 1, 1, 128) 0
flatten_1 (Flatten) (None, 128) 0
dropout_1 (Dropout) (None, 128) 0
dense_1 (Dense) (None, 1024) 132096
dense_2 (Dense) (None, 256) 262400
dense_3 (Dense) (None, 64) 16448
dense_4 (Dense) (None, 1) 65
Total params: 697,441 Trainable params: 697,441 Non-trainable params: 0
Did u try to run the AutopilotV2.py? Is the steering angle turning or still static?
On Fri, Nov 22, 2019, 7:29 PM maxmstrmn [email protected] wrote:
It seems the training is crazy. I printed both model summaries using the print_summary method from the keras utils.
Here ist the output:
`summary repo Model: "sequential_1"
Layer (type) Output Shape Param #
lambda_1 (Lambda) (None, 100, 100, 1) 0
conv2d_1 (Conv2D) (None, 100, 100, 32) 320
activation_1 (Activation) (None, 100, 100, 32) 0
max_pooling2d_1 (MaxPooling2 (None, 50, 50, 32) 0
conv2d_2 (Conv2D) (None, 50, 50, 64) 18496
activation_2 (Activation) (None, 50, 50, 64) 0
max_pooling2d_2 (MaxPooling2 (None, 25, 25, 64) 0
conv2d_3 (Conv2D) (None, 25, 25, 128) 73856
activation_3 (Activation) (None, 25, 25, 128) 0
max_pooling2d_3 (MaxPooling2 (None, 12, 12, 128) 0
flatten_1 (Flatten) (None, 18432) 0
dropout_1 (Dropout) (None, 18432) 0
dense_1 (Dense) (None, 128) 2359424
dense_2 (Dense) (None, 64) 8256
dense_3 (Dense) (None, 1) 65
Total params: 2,460,417 Trainable params: 2,460,417 Non-trainable params: 0
summary my own Model: "sequential_1"
Layer (type) Output Shape Param #
lambda_1 (Lambda) (None, 100, 100, 1) 0
conv2d_1 (Conv2D) (None, 100, 100, 32) 320
activation_1 (Activation) (None, 100, 100, 32) 0
max_pooling2d_1 (MaxPooling2 (None, 50, 50, 32) 0
conv2d_2 (Conv2D) (None, 50, 50, 32) 9248
activation_2 (Activation) (None, 50, 50, 32) 0
max_pooling2d_2 (MaxPooling2 (None, 25, 25, 32) 0
conv2d_3 (Conv2D) (None, 25, 25, 64) 18496
activation_3 (Activation) (None, 25, 25, 64) 0
max_pooling2d_3 (MaxPooling2 (None, 12, 12, 64) 0
conv2d_4 (Conv2D) (None, 12, 12, 64) 36928
activation_4 (Activation) (None, 12, 12, 64) 0
max_pooling2d_4 (MaxPooling2 (None, 6, 6, 64) 0
conv2d_5 (Conv2D) (None, 6, 6, 128) 73856
activation_5 (Activation) (None, 6, 6, 128) 0
max_pooling2d_5 (MaxPooling2 (None, 3, 3, 128) 0
conv2d_6 (Conv2D) (None, 3, 3, 128) 147584
activation_6 (Activation) (None, 3, 3, 128) 0
max_pooling2d_6 (MaxPooling2 (None, 1, 1, 128) 0
flatten_1 (Flatten) (None, 128) 0
dropout_1 (Dropout) (None, 128) 0
dense_1 (Dense) (None, 1024) 132096
dense_2 (Dense) (None, 256) 262400
dense_3 (Dense) (None, 64) 16448
dense_4 (Dense) (None, 1) 65
Total params: 697,441 Trainable params: 697,441 Non-trainable params: 0
Process finished with exit code 0 `
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBF4KKPAF62QNNYDLTLQU67DPA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5LZXI#issuecomment-557497565, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBDWICR474JWBCS5SILQU67DPANCNFSM4HI6765Q .
Everything is tested with the v2 files. The angle is still static. Please note the remarkable difference between these two models I posted above.
If I use the model .h5 shipped in this repo everything works fine. But my own trained fails.
Yes true. I just realised the new model trained is only 8.258 mb whereas the real model is 28.89 mb. Something wrong somewhere
On Fri, Nov 22, 2019, 8:04 PM maxmstrmn [email protected] wrote:
Everything is tested with the v2 files. The angle is still static. Please note the remarkable difference between these two models I posted above.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBE27AG4ZGWA5SU65IDQU7DFBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5OEFQ#issuecomment-557507094, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBEJPI2CWURZOREXYS3QU7DFBANCNFSM4HI6765Q .
However, I have tested with own dataset and the steering angle is moving. It's weird but working
On Fri, Nov 22, 2019, 10:16 PM Sriram Ramasamy [email protected] wrote:
Yes true. I just realised the new model trained is only 8.258 mb whereas the real model is 28.89 mb. Something wrong somewhere
On Fri, Nov 22, 2019, 8:04 PM maxmstrmn [email protected] wrote:
Everything is tested with the v2 files. The angle is still static. Please note the remarkable difference between these two models I posted above.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/akshaybahadur21/Autopilot/issues/3?email_source=notifications&email_token=AFBSJBE27AG4ZGWA5SU65IDQU7DFBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5OEFQ#issuecomment-557507094, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBSJBEJPI2CWURZOREXYS3QU7DFBANCNFSM4HI6765Q .
could you please share the whole code for AutopilotV2.py. Because when i went through the whole code there was no difference between V1 and V2 code(the Autopilot V1.py and Autopilot V1.py) And even there is no "run.mp4" video for V2 Could you share these stuff...would be really helpful
The run.mp4 for v2 is not uploaded due to copyright issues. Use video from youtube
