Supporting image compression formats passthrough (#1461)

* Supporting ros depth compression formats passthrough

* Added image_compression_format (support jpg,png). Use Compression.h constants.

* Bridging gap with #1457

* CoreWrapper: decode_images_on_demand

* rgbd_sync->rgbd_split encoding/decoding test with image_transport validation. rgbd_split color passthrough

* more tests

* fixing ci on kilted+

* fixing test

* fixing lyrical tests

* fixing rolling ci

* Improved test coverage

* fixing some tf waitForTransform deadlocks when sim time is used in tests

* CI: build rtabmap_slam on rolling

rtabmap_slam builds without nav2_msgs, which is not on resolute yet: skip its
key instead of leaving the package out. Only rtabmap_costmap_plugins (and the
rtabmap_ros metapackage pulling it in) still need nav2.

rtabmap_sync/doc/rgbd_sync.md

@@ -111,7 +111,7 @@ ComposableNode(
 | Topic | Type | Description |
 |---|---|---|
 | `rgbd_image` | [`rtabmap_msgs/msg/RGBDImage`](https://docs.ros.org/en/jazzy/p/rtabmap_msgs/msg/RGBDImage.html) | The three inputs, raw. Published only when someone is subscribed. |
-| `rgbd_image/compressed` | [`rtabmap_msgs/msg/RGBDImage`](https://docs.ros.org/en/jazzy/p/rtabmap_msgs/msg/RGBDImage.html) | The same frame with JPEG color and PNG depth instead of raw images. Published only when someone is subscribed. |
+| `rgbd_image/compressed` | [`rtabmap_msgs/msg/RGBDImage`](https://docs.ros.org/en/jazzy/p/rtabmap_msgs/msg/RGBDImage.html) | The same frame with compressed color (JPEG by default) and depth (PNG by default) instead of raw images. Published only when someone is subscribed. |
 
 The output's `header.frame_id` is taken from the **camera_info**, which is the frame the calibration is expressed in — so make sure the images and the `camera_info` carry the same `frame_id`. If they disagree, the output is labelled with the calibration's frame while the pixels were measured in another, and every point projected out of them lands somewhere else.
 
@@ -132,6 +132,8 @@ The output's `header.stamp` is the **later** of the color and depth stamps, so t
 | `depth_scale` | `double` | `1.0` | Multiplies every depth pixel. See [Depth units](#depth-units). |
 | `decimation` | `int` | `1` | Downsample both images by this factor, scaling the calibration to match. Must divide the depth image size exactly, or it is ignored. |
 | `compressed_rate` | `double` | `0.0` | Maximum rate, in Hz, of `rgbd_image/compressed`. `0` means every frame. Does not affect `rgbd_image`. |
+| `image_compression_format` | `string` | `".jpg"` | Format of the color image in `rgbd_image/compressed`: `".jpg"` (lossy, smaller) or `".png"` (lossless, larger). An invalid value falls back to `".jpg"` with an error. |
+| `depth_compression_format` | `string` | `".png"` | Format of the depth in `rgbd_image/compressed`: `".png"`, `".rvl"`, `".png:<maxDepth>:<quantization>"`, `".rvl:<maxDepth>:<quantization>"`, any of them prefixed by `"legacy:"`, or `"legacy"`. See [Compressing for a slow link](#compressing-for-a-slow-link). |
 | `image_transport` | `string` | `"raw"` | Transport for `rgb/image`, e.g. `compressed`. |
 | `depth_transport` | `string` | `"raw"` | Transport for `depth/image`, e.g. `compressedDepth`. |
 | `rgb_image_transport` | `string` | — | **Deprecated**, renamed to `image_transport`. |
@@ -160,7 +162,24 @@ RTAB-Map reads `16UC1` depth as millimeters and `32FC1` as meters. A driver that
 
 ## Compressing for a slow link
 
-`rgbd_image/compressed` carries the same frame with the color image as **JPEG** and the depth image as **PNG**. Depth stays lossless deliberately: JPEG artifacts in a depth image are not blur, they are invented geometry.
+`rgbd_image/compressed` carries the same frame with the color image as **JPEG** and the depth image as **PNG**. Depth stays lossless deliberately: JPEG artifacts in a depth image are not blur, they are invented geometry. `image_compression_format: ".png"` makes the color image lossless too, at a few times the size.
+
+The images are in the formats of `compressed_image_transport` (e.g. `"bgr8; jpeg compressed bgr8"`) and `compressed_depth_image_transport` (e.g. `"16UC1; compressedDepth png"`), so the `compressed` and `compressedDepth` image_transport plugins can read them.
+
+`depth_compression_format` changes the depth format:
+
+| Value | 16UC1 depth | 32FC1 depth |
+|---|---|---|
+| `".png"` (default) | Lossless PNG | Rounded to millimeters (16UC1, ±0.5 mm, depth over 65.5 m lost), PNG |
+| `".rvl"` | Lossless RVL: several times faster to compress than PNG, but larger. PNG before Jazzy, as `compressed_depth_image_transport` cannot decode RVL there. | Millimeters, RVL |
+| `".png:<maxDepth>:<quantization>"`, `".rvl:<maxDepth>:<quantization>"` | Same as without the depth parameters | Quantized on 16 bits as inverse depth, like `compressed_depth_image_transport` (e.g. `".png:10:100"`, its default): smaller, stays 32FC1, but depth over `maxDepth` is lost and the error grows with the square of the depth (~0.05 mm at 1 m and 5 mm at 10 m with a quantization of 100). |
+| `"legacy:<format>"` (`"legacy"` is `"legacy:.png"`) | With the codec | Without depth parameters: **lossless**, 4 channels PNG of the float bytes, whatever the codec. With depth parameters: inverse depth. |
+
+`"legacy:"` keeps rtabmap's own format, the one of its database (`Mem/DepthCompressionFormat`) and of rtabmap_ros before 0.24, instead of `compressed_depth_image_transport`'s. image_transport plugins cannot read it (format `"png"`, `"rvl"` or `"rvl:10:100"`), but rtabmap_ros decodes it, and [rgbd_split](https://docs.ros.org/en/jazzy/p/rtabmap_util/)'s `compressedDepth` output converts it. Use it to:
+
+- keep 32FC1 depth lossless when compressed (`"legacy"`);
+- send RVL before Jazzy without it being re-compressed as PNG (`"legacy:.rvl"`);
+- feed rtabmap_ros nodes older than 0.24 (`"legacy"` or `"legacy:.rvl"`; they cannot decode inverse depth).
 
 Neither output is produced unless it has a subscriber, so the compression costs nothing until something subscribes.