Open yanpanlau opened 8 years ago
Hi yanpanlau,
Did you correctly set the window size as explained in the readme (the process below)?
My implementation only supports 64x64 visual input. The vision is captured by simply copying the actual application window, as the original torcs does.
So what you seeing in the matplotlib window will be the left-bottom corner of the second picture...
Great. Now it is working properly. Thanks for the prompt reply.
Although set to 64x64 works, but it is very hard to visualize the RL agent. Can I ask
1) How can I set the GUI size to say 1024x768 but keep the image output as 64x64 2) How can one set default display car velocity on screen? Currently I need to press "0" everytime 3) How can one set default to turn off the engine sound?
Thanks
1) How can I set the GUI size to say 1024x768 but keep the image output as 64x64
In our code, we cannot do it.
2) How can one set default display car velocity on screen? Currently I need to press "0" everytime
hmmm..., I remember that I didn't set these parameters (?) in my experiment. I usually drove the car in the practice mode. At that time, it looked no problem. So could you explain the detail of your situation?
3) How can one set default to turn off the engine sound?
Because I am currently apart from my ubuntu machine, I cannot try torcs on my machine right now (I'm replying by my macbook;) ). But there may be some option in torcs to turn of the sound (sorry, actually I am not sure right now). You may need to run the torcs by super-user to save the configuration.
To turn off the engine sound and to display the velocity you can either modify the "practice.xml" file with the proper options or simply adapt the "autostart.sh" file to your needs
Hi, how did you set the resolution to 64x64? On my computer the lowest resolution 320x240 and doesn't 64x64 too small as vision input of deep neuron network, e.g. CNNs?
If you cannot set the screen resolution with options then change it in
change x->64 and y->64
I keep getting either 0 for all pixels and rgb or 255. I have the display set to 64. Any ideas?
Same as eubrunomiguel, I configured it many times to be sure, screen resolution is 64*64, but all I get in the images are zeros.
Hello @ugo-nama-kun and @yanpanlau I am trying to train an agent that gets as input the raw pixels of the game screen. Your implementation allows for getting visual input as an observation so I find it great.
Nevertheless, when I try to set the window size as explained in the process below the game restarts and the window size is again 640 x 480.
"After the selection of the driving-window mode, you need to set the appropriate gui size. This is done by using the display option mode in Options --> Display. You can select the Screen Resolution, and you need to select 64x64 for visual input (our immplementation only support this screen size, other screen size results the unreasonable visual information)."
Is there something I am missing here?
Edit : I forgot to mention that I work in a linux server via SSH and I get the game screen by enabling X11 forwarding.
I changed the resolution to 64x64, but the extracted image is completely black, does anyone know why?@ugo-nama-kun
I changed the resolution to 64x64, but the extracted image is completely black, does anyone know why?@ugo-nama-kun
Same... client.S.d.get('img)
is just full of 0s
I changed the resolution to 64x64, but the extracted image is completely black, does anyone know why?@ugo-nama-kun
Same...
client.S.d.get('img)
is just full of 0s
Hello,I have solved my problem,the reason for my problem is that the graphics card driver is not applied. When you install the graphics driver,cuda and cudnn,you still need to check whether your system uses your discrete graphics driver. You can check it in the system details,hope to help you.
I changed the resolution to 64x64, but the extracted image is completely black, does anyone know why?@ugo-nama-kun
Same...
client.S.d.get('img)
is just full of 0sHello,I have solved my problem,the reason for my problem is that the graphics card driver is not applied. When you install the graphics driver,cuda and cudnn,you still need to check whether your system uses your discrete graphics driver. You can check it in the system details,hope to help you.
Hey thank you for your message. I do not think that is the issue. It works fine on the remote machine where I run the code, just not through a SHH connection, even tho X11 is enabled and it is rendered correctly. Did you also run it through SSH?
Hi, I am running example_experiment.py and I try to plot the visual input in sample_agent.py. However, I didn't get the proper image. Attached is the screenshot. Can you please advise? Thanks
The Game Console is running properly and I can see the car is running