Fix instance detection ros image

[RobinSchmid7]

The instance() method does not work (on my system) with the ros_img.
Replaced by inbuilt type + .__name__ method which works.

Output of

wild_visual_navigation/wild_visual_navigation_ros/src/wild_visual_navigation_ros/ros_converter.py

Lines 108 to 117 in 989e56a

	def ros_image_to_torch(ros_img, desired_encoding="rgb8", device="cpu"):
	if type(ros_img).__name__ is "_sensor_msgs__Image":
	np_image = CV_BRIDGE.imgmsg_to_cv2(ros_img, desired_encoding=desired_encoding)
	elif type(ros_img).__name__ is "_sensor_msgs__CompressedImage":
	np_arr = np.fromstring(ros_img.data, np.uint8)
	np_image = cv2.imdecode(np_arr, cv2.IMREAD_COLOR)
	if "bgr" in ros_img.format:
	np_image = cv2.cvtColor(np_image, cv2.COLOR_BGR2RGB)
	return TO_TENSOR(np_image).to(device)

for ros_img :

print(isinstance(ros_img, Image)): False
print(isinstance(ros_img, CompressedImage)): False
print(type(ros_img).__name__ is "_sensor_msgs__Image"): True

[mmattamala]

isinstance should be fairly standard. Can you also try to compare the type of ros_msg with the type of Image for instance?

[RobinSchmid7]

Yes, it seems to be a type difference:

a = Image()
print(type(a)): <class 'sensor_msgs.msg._Image.Image'>
print(type(ros_msg): <class 'tmpjz2xjmha._sensor_msgs__Image'>

So the tmpjz... refers to a unique identifier which changes each time I run the code. The actual python class for the sensor_msgs seems to be dynamically allocated. Maybe we can we fix this dynamic allocation?

[RobinSchmid7]

Problem resolved: the problem only occurs if bags are replayed using the BagTfTransformer package from python not when replaying via a terminal. This generates a dynamically allocated python class for the sensor_msgs topics which is not recognized with the isinstance() method. Since the ros_image_to_torch() method is used both during deployment and offline data generation I added both options so its compatible.