Steering not turning for the Autopilotv2 and i am using the same run video and same code

[vineethamurali]

[maxmstrmn]

Same here. Have you solved your problem?

[maxmstrmn]

Did you succeed in the meantime?

[Sriram97]

Yes ita working now.

…

On Tue, Nov 5, 2019, 3:37 PM maxmstrmn ***@***.***> wrote: Did you succeed in the meantime? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#3?email_source=notifications&email_token=AFBSJBC5WRMQXEH7UAMVXJ3QSEPEBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDB4EXQ#issuecomment-549700190>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBSJBAB7JVF3CCWCAOQPNLQSEPEBANCNFSM4HI6765Q> .

[maxmstrmn]

What did you do? Could you share the source code?

[Sriram97]

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 ***@***.***> 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 <#3?email_source=notifications&email_token=AFBSJBFB6LE754LBXTDDEJTQSVVMDA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDSDZZI#issuecomment-551828709>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBSJBHCNTTCRCR3KFMYMALQSVVMDANCNFSM4HI6765Q> .

[maxmstrmn]

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?

[Sriram97]

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 ***@***.***> 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 <#3?email_source=notifications&email_token=AFBSJBDVJ3FJPCVCJFPLQL3QUZLLBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEEZUJ6Q#issuecomment-557008122>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBSJBCLPQOJ7BOQG5NVQLTQUZLLBANCNFSM4HI6765Q> .

[maxmstrmn]

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.

[maxmstrmn]

No success, even with the larger dataset the final .h5 file is around 8mb large.

[maxmstrmn]

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

---

[Sriram97]

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 ***@***.***> 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 <#3?email_source=notifications&email_token=AFBSJBF4KKPAF62QNNYDLTLQU67DPA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5LZXI#issuecomment-557497565>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBSJBDWICR474JWBCS5SILQU67DPANCNFSM4HI6765Q> .

[maxmstrmn]

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.

[Sriram97]

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 ***@***.***> 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 <#3?email_source=notifications&email_token=AFBSJBE27AG4ZGWA5SU65IDQU7DFBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5OEFQ#issuecomment-557507094>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBSJBEJPI2CWURZOREXYS3QU7DFBANCNFSM4HI6765Q> .

[Sriram97]

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 ***@***.***> 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 ***@***.***> 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 > <#3?email_source=notifications&email_token=AFBSJBE27AG4ZGWA5SU65IDQU7DFBA5CNFSM4HI67652YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE5OEFQ#issuecomment-557507094>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AFBSJBEJPI2CWURZOREXYS3QU7DFBANCNFSM4HI6765Q> > . >