carla-simulator / carla

Open-source simulator for autonomous driving research.
http://carla.org
MIT License
10.83k stars 3.48k forks source link

Distorted and Transparent Objects #4761

Closed mrcoee closed 2 years ago

mrcoee commented 2 years ago

I currently would like to generate a few thousand images of random traffic scenes. For this purpose I randomly spawn pedestrians and vehicles, make an rgb and an sematic segmantation image, and despwan every spawned objects afterwards. Unfortunately I occasionally get partially/fully distored vehicles and transparent pedestrians:

1_9

The RGB sensor attributes are: image_size_x :1920 image_size_x: 1080 fov: 110 sensor_tick: 1

I'm using Carla 0.9.11 and 0.9.12 on Ubuntu18.04.

MattRowe18 commented 2 years ago

@mrcoee it may be that the camera catches the vehicles or the pedestrians as they are being spawned or destroyed. Some operations CARLA are multithreaded and asynchronous, including the operations that render sensor images, therefore one thread may be spawning the vehicle while another thread samples the sensor image.

Are you using CARLA in synchronous mode? If not, this could help solve the problem. If you are already using synchronous mode, I would then suggest dividing your process into spawn cycles and image cycles. i.e. spawn hundreds of vehicles and pedestrians, tick() numerous times, then randomly place your cameras and take the images, then rinse and repeat.

If the problem persists, please reopen this issue.

yuvalHG commented 2 years ago

@MattRowe18 I work with synchronous mode (Carla 9.11), first spawning all actors and later after applying auto pilot to each, snap images from RGB camera on the Ego vehicle and same problem is observed.

It seems that this issue of fuzzy images / not fully rendered vehicles is repeatedly reported, here are some examples:

  1. [https://github.com/carla-simulator/carla/issues/4860](url)
  2. https://github.com/carla-simulator/carla/issues/4764#issue-1029917913

Is there any walkaround for this issue? Is it due to improper chose of camera parameters?

Thanks Yuval

wuhanstudio commented 7 months ago

Similar issue using Carla 0.9.14.

Two vehicles on the left side are blurry.

# Create a RGB camera
camera_bp = bp_lib.find('sensor.camera.rgb')

camera_bp.set_attribute('image_size_x', '1392')
camera_bp.set_attribute('image_size_y', '1024')

# 72 degrees # Always fov on width even if width is different than height
camera_bp.set_attribute('fov', '72')
camera_bp.set_attribute('enable_postprocess_effects', 'True')
camera_bp.set_attribute('sensor_tick', '0.10')  # 10Hz camera
camera_bp.set_attribute('gamma', '2.2')
camera_bp.set_attribute('motion_blur_intensity', '0')
camera_bp.set_attribute('motion_blur_max_distortion', '0')
camera_bp.set_attribute('motion_blur_min_object_screen_size', '0')
camera_bp.set_attribute('shutter_speed', '1000')  # 1 ms shutter_speed
camera_bp.set_attribute('lens_k', '0')
camera_bp.set_attribute('lens_kcube', '0')
camera_bp.set_attribute('lens_x_size', '0')
camera_bp.set_attribute('lens_y_size', '0')

0020

If I set camera_bp.set_attribute('enable_postprocess_effects', 'False):

0016