You added a Camera display in RViz, picked your topic from the list, and the panel stays grey. The status block stacks up three complaints: No tf data, Fixed Frame [/camera] does not exist, and CameraInfo: No CameraInfo received. This is the standard first-hour fault when a custom node publishes images through image_transport and nothing renders, and it catches people out because the topic selection worked. This page is for engineers debugging ROS and ROS 2 camera streams. You will read the status block, fix each missing input in order of probability, and decide when the plain Image display is the better tool.
The three inputs the Camera display waits for
The Camera display does more than paint pixels. It draws the image as a texture on a viewing frustum placed inside the 3D scene, and it refuses to draw until three inputs arrive together:
- an image topic carrying
sensor_msgs/msg/Image(sensor_msgs/Imagein ROS 1) - a calibration message,
sensor_msgs/msg/CameraInfo, on the paired info topic: current rviz2 builds expect it at<image topic>/camera_info, while the image_transport camera publisher places it at<base name>/camera_info - a TF transform from the global Fixed Frame to the
frame_idin the message headers, evaluated at the image's own timestamp
The intrinsics inside CameraInfo, focal lengths and principal point, set the frustum geometry, and TF sets its pose. Miss any one of the three and the display draws nothing, but each miss writes its own line in the status block, so the block is your diagnosis, not a generic error.
Read the status block before touching anything
The display reports each missing input separately. Work the lines in this order:
| Status line | Missing input | Test | First fix |
|---|---|---|---|
| No tf data; Fixed Frame [/camera] does not exist | TF into the fixed frame | echo the image header's frame_id; open the TF tree |
set Fixed Frame to that frame_id, or publish the camera pose statically |
| No CameraInfo received on [/camera_info] |
sensor_msgs/msg/CameraInfo on the paired name |
list topics and filter for camera_info | publish calibration, pair the names by remap, or use an Image display |
| No image received |
sensor_msgs/msg/Image arriving on the chosen topic |
ros2 topic info and ros2 topic hz
|
fix the publisher, the message type or the QoS |
TF comes first because a missing fixed frame blocks every display in the window, not just the Camera panel. The middle row is the one hand-written publisher nodes hit most, and the next sections take the causes in the order you will meet them.
Cause 1: the fixed frame is not a frame anyone publishes
Setting Fixed Frame to /camera feels natural when your topic is /camera/image_raw, but the fixed frame is not a topic name. It is a coordinate frame in the TF tree, and no camera driver publishes a frame called /camera. That one assumption produces the whole first line of the failure: No tf data, then Fixed Frame [/camera] does not exist.
- Set Fixed Frame to the
frame_idyour image header actually carries. Read it withros2 topic echo /camera/image_raw --field header --once. When the fixed frame equals the message's own frame, the transform is the identity and every display in the window comes alive at once. - Keep Fixed Frame on
maporbase_linkand publish the camera pose once, statically, with thestatic_transform_publishernode fromtf2_ros, parent your fixed frame, child the camera frame. The frustum then sits in the scene at the camera's true pose, which is the point of the Camera display.
If the window still reports the frame missing while your driver runs, the frame name in the TF tree and the header's frame_id differ by a slash or a suffix; the fixed frame walkthrough walks that comparison line by line.
Cause 2: the node publishes Image and nobody publishes CameraInfo
An image_transport::Publisher sends sensor_msgs/Image and nothing else. It has no calibration to attach, so the warning No CameraInfo received on [/camera_info] is literally accurate: it names a message no node produces. Hand-written publisher nodes cause almost every instance of this fault; drivers for area scan cameras publish the image and its calibration together.
Check what exists with ros2 topic list | grep camera_info (rostopic list in ROS 1). Empty output means no calibration anywhere on the system, and the display will not invent it. Then pick by intent:
- Debugging aid: publish CameraInfo for real. Put the intrinsics in the standard calibration YAML, load it with a camera info manager node, and let it republish the message with every frame. CameraInfo carries width and height, the distortion coefficients, and the 3 × 3 matrix K with focal lengths and principal point.
- You want the picture: swap the Camera display for a plain Image display on the same topic. The Image display subscribes to the image alone and never asks for TF or calibration.
Cause 3: CameraInfo exists but the topic names do not pair
Some drivers publish calibration under a layout the display does not build. A common multi-stream layout puts it at <stream>/camera_info, for example /camera/color/camera_info, while an rviz2 Camera display pointed at /camera/color/image_raw expects /camera/color/image_raw/camera_info. Both messages exist, the pairing misses, and the display keeps reporting No CameraInfo received even though a working calibration is one remap away.
List every calibration topic first, then make the names agree: remap the driver's camera_info output to the name the display derives, or point the display's info topic property at the published one where your rviz2 build offers it. Echo one CameraInfo message and check that its header frame_id equals the image's; a mismatch there breaks the frustum placement even when the topics pair. Stereo and ToF drivers in the 3D sensors category publish one calibration per stream, so expect several camera_info topics and pick the one for the stream you subscribed to. When the topic itself is hard to find, the topic-name walkthrough shows how to list what a node really publishes.
Cause 4: the topic carries the wrong type or only compressed data
ros2 topic info /camera/image_raw prints the message type on the topic, and it is the fastest sanity check in this debug. The image half of a Camera display, like the Image display, subscribes to sensor_msgs/msg/Image; a topic typed sensor_msgs/msg/CompressedImage reports as No image received on both. Compressed-only streams are the usual trap: image_transport publishes /camera/image_raw/compressed beside the raw name, but a display needs something serving the raw name to hand it frames. If your node emits only compressed data, either feed a compressed display or add the republish step. The compressed streams article covers that chain, and the transport compression parameters article explains which transport topics exist in the first place.
Cause 5: ROS 2 QoS keeps the subscription from matching
Bytes can leave the publisher at the right rate and still never reach RViz when the quality-of-service profiles disagree. Camera topics are sensor data and are usually published best-effort; a display asking for reliable never matches one, and no warning explains why. Set the display's Reliability setting to Best Effort in rviz2, or make the publisher reliable. Either side may yield. The symptom is distinctive: ros2 topic hz shows the full frame rate in one terminal while the display keeps saying No image received. On depth streams the same mismatch shows as flicker rather than blank, which the depth image QoS article traces step by step.
Worked example: an image_transport node publishing /camera/image_raw only
Every input, named and sourced:
- Publisher: a custom node using
image_transport::Publisheron/camera/image_raw(the set-up behind the original fault) - Format: 640 × 480,
rgb8, 3 bytes per pixel (set in the node's code) - Target rate: 30 Hz (the loop rate given to the node)
- RViz status: No tf data; Fixed Frame [/camera] does not exist; No CameraInfo received on [/camera_info]; No image received (read from the display panel)
Rule: prove the bytes leave the node before touching RViz. A frame's size is width × height × bytes per pixel, and the wire load is that size times the rate times 8 bits per byte:
- frame: 640 × 480 × 3 = 921,600 bytes
- rate: 921,600 bytes × 30 Hz = 27,648,000 bytes/s
- wire: 27,648,000 bytes/s × 8 = 221,184,000 bit/s ≈ 221 Mbit/s
Measured, ros2 topic hz /camera/image_raw reports 30.0 Hz, and 221 Mbit/s sits well inside a 1,000 Mbit/s Ethernet link, so the publisher streams at its set rate and the network is not the limit.
Decision, with its threshold: at the set frame rate, both remaining complaints are RViz-side and both are real. /camera is a topic name, not a TF frame, so the fixed-frame lookup fails, and a plain image_transport::Publisher never emits CameraInfo. Set Fixed Frame to the header's frame_id, then publish CameraInfo from the calibration file or switch to an Image display on /camera/image_raw. The split: hz at your set rate means fix RViz; hz at 0 means fix the node, where the usual causes are a single publish at start-up instead of a per-frame loop, or a type other than sensor_msgs/msg/Image. A node that publishes one frame before RViz subscribes reproduces exactly this blank state.
When the plain Image display is the right answer
Choose by what the picture is for. The Camera display earns its set-up cost when you want the image drawn inside the 3D scene at the camera's true pose, or when your pipeline rectifies with image_proc, which consumes CameraInfo directly. For exposure, framing and focus checks, the Image display on the raw topic needs no TF, no calibration and no fixed frame, on your desk screen or on a cell-mounted industrial touch monitor.
- Screen size
- 10.1"
One dependency to know: the rectified topic image_rect exists only while CameraInfo is published, because rectification is computed from it. That is why image_raw and image_rect both exist, and why stopping the calibration node silently removes the rectified stream too.
Prevention: what a well-formed camera node publishes
Six habits remove this entire class of fault:
- Image and CameraInfo from the same node, with the same stamp and the same
frame_id, on names that pair - a
frame_idending in_optical, with the pose to it published once as a static transform from your cell or robot frame - a publish call inside the capture loop, so a frame always exists when RViz subscribes
- stamps taken from the node clock every frame, so TF lookups at the image time succeed
- one calibration session, one YAML file, republished by a camera info manager node from then on
-
ros2 topic list,infoandhzchecked before RViz opens, so the display is the last suspect, not the first
The rest of the ROS debugging set sits on the systems topic page.
The decision, case by case
Map each status line to its fix and the display renders on the next frame:
- Fixed Frame does not exist: set the fixed frame to the header's
frame_id, or publish the camera pose statically - No CameraInfo received: publish calibration under the paired name, remap the driver's output to pair, or switch to an Image display
- No image received at full hz: match the message type and the QoS reliability
- No image received at 0 Hz: the node; loop the publish and check the type
When no topics appear at all, work outward from the device: the video device permission checks and the wrist camera video device walkthrough cover the USB and driver layer underneath RViz. When the real fault is a camera with no calibrated driver yet, choose hardware whose driver publishes the image and calibration pair from the start: the camera finder filters by interface and resolution, and every order includes free expert support for integration questions like this one. If the display still refuses after all four fixes, send the status block through the contact page and we will read it with you.
Questions
Why does the Camera display need CameraInfo when the Image display does not?
The Camera display places your image on a frustum inside the 3D scene, and the intrinsics in CameraInfo (focal lengths, principal point, distortion) set that frustum geometry. It also needs TF to position the frustum. The Image display only paints pixels into a panel, so neither calibration nor TF exists for it. Use Camera when the view needs world context, Image when you need the picture itself.
What goes inside a CameraInfo message?
Width and height, the distortion coefficients, and the 3 × 3 intrinsic matrix K holding focal length in pixels and the principal point, plus the projection matrix P. You produce it by calibrating against a checkerboard target, and the result is saved as a YAML file that a camera info manager node loads and republishes with every frame. Until you calibrate, a datasheet-based CameraInfo lets RViz draw, but overlays stay approximate.
Where should camera_info be published for the pairing to work?
At the name the display derives from your image topic. Current rviz2 builds append camera_info to the image topic name, so /camera/image_raw pairs with /camera/image_raw/camera_info, while the image_transport camera publisher convention places it at the base name plus camera_info. Echo your CameraInfo topic, compare both constructions against it, and remap whichever side you control until the names agree.
The image shows once and then never updates: what causes that?
Timestamps. The Camera display looks up TF at each image's header.stamp, so zero stamps, one stamp reused forever, or simulation time with no clock publisher make every lookup after the first fail. Stamp each message from the node clock inside the capture loop, and run your TF publishers on the same clock source. The Image display keeps painting regardless, which is a quick way to confirm the stream itself is alive.
Does a compressed stream work with the Camera display?
Only if something also serves the raw image name. The display subscribes through image_transport to the base topic, and image_transport serves the compressed variant through its own republish chain; a node emitting only sensor_msgs/msg/CompressedImage leaves that chain empty. Publish raw and let the transport compress, or run the republish node, and check ros2 topic info to confirm the type on the topic you selected.
