================================================================================ Huhb3D - ROS image_transport / cv_bridge 16UC1 Compliance Report ================================================================================ === 1. PNG FILE FORMAT INSPECTION === Test file: depth_0001.png (1510 bytes) PNG signature: 89504e470d0a1a0a Valid PNG: NO IHDR chunk analysis: Width: 800 Height: 600 Bit depth: 16 Color type: 0 (Grayscale) Compression: 0 Filter method: 0 Interlace: 0 ROS 16UC1 Requirements Check: Color type == 0 (Grayscale): PASS (actual: 0/Grayscale) Bit depth == 16: PASS (actual: 16) No interlacing: PASS (actual: 0) [OK] PNG format is EXACTLY 16-bit Grayscale - fully compatible with ROS 16UC1 Critical: cv_bridge expects single-channel 16-bit unsigned. If PNG is saved as 8-bit grayscale or RGB, cv_bridge will produce wrong encoding or black image. === 2. cv_bridge SIMULATION === cv2.imread(IMREAD_UNCHANGED): shape: (600, 800) dtype: uint16 ndim: 2 cv_bridge conversion checks: ndim == 2 (single channel): PASS (actual: 2) dtype == uint16: PASS (actual: uint16) shape == (600, 800): PASS (actual: (600, 800)) [OK] cv_bridge will correctly encode as sensor_msgs/Image with: encoding: '16UC1' height: 600 width: 800 step: 1600 (width * sizeof(uint16)) cv_bridge Python code: from cv_bridge import CvBridge bridge = CvBridge() depth = cv2.imread('depth_0001.png', cv2.IMREAD_UNCHANGED) msg = bridge.cv2_to_imgmsg(depth, encoding='16UC1') # msg.encoding == '16UC1' # msg.step == 1600 # msg.data == depth.tobytes() Common cv_bridge pitfall simulation: cv2.imread(IMREAD_COLOR): shape=(600, 800, 3), dtype=uint8 [WARN] If user forgets IMREAD_UNCHANGED, gets 8UC3 BGR -> BLACK IMAGE! Correct: cv2.imread(path, cv2.IMREAD_UNCHANGED) cv2.imread(IMREAD_GRAYSCALE): shape=(600, 800), dtype=uint8 Range: [2, 3] (WRONG! High byte lost!) [WARN] IMREAD_GRAYSCALE converts to 8-bit -> loses high byte -> WRONG VALUES! === 3. image_transport / sensor_msgs/Image BYTE-LEVEL VERIFICATION === sensor_msgs/Image binary layout: encoding: '16UC1' is_bigendian: 0 (x86 little-endian) step: 1600 data length: 960000 bytes First 32 bytes (row 0, pixels 0-15): 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 Byte order verification (little-endian uint16): Pixel [0,0]: value=0, bytes=[0x00, 0x00], reconstructed=0 Pixel [0,1]: value=0, bytes=[0x00, 0x00], reconstructed=0 Little-endian match: PASS Step (stride) verification: step = 1600 bytes/row Pixel [1,0]: value=0, reconstructed=0 Stride alignment: PASS image_transport compressed transport check: PNG compression is lossless -> depth values preserved exactly For compressed_depth transport, use depth_image_proc package: ros2 run depth_image_proc depth_image_proc Input: sensor_msgs/Image encoding='16UC1' Output: sensor_msgs/Image encoding='32FC1' (float, meters) Conversion: float_value = uint16_value / 1000.0 === 4. DEPTH VALUE RANGE vs ROS 16UC1 LIMITS === uint16 range: 0 to 65535 Our depth range: ~730 to ~844 mm Max representable: 65535 mm = 65.535 m Value distribution check: Total non-zero pixels: 45440 Min value: 730 (0x02da) Max value: 844 (0x034c) Mean value: 787.4 Std value: 22.0 Overflow (>65535): 0 PASS Underflow (<0): 0 PASS [OK] All values fit within uint16 range [OK] No overflow/underflow in ROS 16UC1 encoding === 5. COMMON ROS DEPLOYMENT PITFALLS CHECK === Pitfall 1: Wrong cv2.imread flag WRONG: cv2.imread(path) -> 8UC3 BGR, black image WRONG: cv2.imread(path, IMREAD_GRAYSCALE) -> 8UC1, high byte lost RIGHT: cv2.imread(path, IMREAD_UNCHANGED) -> 16UC1, correct values IMREAD_UNCHANGED: dtype=uint16, range=[750, 795] IMREAD_COLOR: dtype=uint8, range=[2, 3] WRONG! IMREAD_GRAYSCALE: dtype=uint8, range=[2, 3] WRONG! (high byte lost) Gray vs Unchanged error: 99.7% (CRITICAL!) Pitfall 2: depth_scale mismatch Our depth_scale = 1.0 (depth PNG stores mm directly) BOP convention: depth_mm = pixel_value * depth_scale ROS convention: depth_m = pixel_value / 1000.0 [OK] depth_scale=1.0 is unambiguous Pitfall 3: Endianness in sensor_msgs/Image is_bigendian field: must be 0 on x86/x64 systems Our data: little-endian (x86 native) cv_bridge sets is_bigendian=0 automatically [OK] No endianness mismatch Pitfall 4: Step/stride alignment step = width * sizeof(uint16) = 800 * 2 = 1600 No padding bytes between rows cv_bridge uses contiguous numpy array (no gaps) [OK] No stride alignment issues Pitfall 5: Zero-value semantics In our data: 0 means 'no depth' (background) In ROS: 0 typically means 'invalid depth' (same semantics) depth_image_proc handles 0 as NaN in float conversion [OK] Zero-value semantics match ROS convention Pitfall 6: image_transport compressed format For compressed_depth transport, PNG is re-encoded 16-bit PNG is lossless -> no data corruption [OK] Lossless compression preserves depth values === 6. cv_bridge ROUND-TRIP TEST === Original shape: (600, 800), dtype: uint16 Roundtrip shape: (600, 800), dtype: uint16 Byte-exact match: PASS Simulated sensor_msgs/Image -> cv_bridge -> cv2 roundtrip: 1. depth_png = cv2.imread(path, IMREAD_UNCHANGED) # uint16 (600,800) 2. msg = bridge.cv2_to_imgmsg(depth_png, '16UC1') 3. depth_back = bridge.imgmsg_to_cv2(msg, '16UC1') 4. assert np.array_equal(depth_png, depth_back) # True [OK] Roundtrip is lossless === 7. ALL OBJECTS DEPTH PNG FORMAT VERIFICATION === bearing_block: [PASS] shape=(600, 800) dtype=uint16 range=[770,801] bearing_medium: [PASS] shape=(600, 800) dtype=uint16 range=[770,803] bearing_small: [PASS] shape=(600, 800) dtype=uint16 range=[779,801] connector_housing: [PASS] shape=(600, 800) dtype=uint16 range=[778,805] coupling: [PASS] shape=(600, 800) dtype=uint16 range=[775,796] flange: [PASS] shape=(600, 800) dtype=uint16 range=[750,795] flange_large: [PASS] shape=(600, 800) dtype=uint16 range=[730,795] flange_medium: [PASS] shape=(600, 800) dtype=uint16 range=[750,797] flange_small: [PASS] shape=(600, 800) dtype=uint16 range=[765,812] gear: [PASS] shape=(600, 800) dtype=uint16 range=[773,797] gear_large: [PASS] shape=(600, 800) dtype=uint16 range=[752,794] gear_medium: [PASS] shape=(600, 800) dtype=uint16 range=[773,796] gear_small: [PASS] shape=(600, 800) dtype=uint16 range=[782,810] heat_sink: [PASS] shape=(600, 800) dtype=uint16 range=[754,844] hex_bolt: [PASS] shape=(600, 800) dtype=uint16 range=[791,799] l_bracket: [PASS] shape=(600, 800) dtype=uint16 range=[784,812] mounting_plate: [PASS] shape=(600, 800) dtype=uint16 range=[745,806] pipe_tee: [PASS] shape=(600, 800) dtype=uint16 range=[776,811] step_shaft: [PASS] shape=(600, 800) dtype=uint16 range=[780,797] valve_body: [PASS] shape=(600, 800) dtype=uint16 range=[759,838] ================================================================================ FINAL VERDICT ================================================================================ ALL CHECKS PASSED - Depth maps are fully ROS 16UC1 compliant Deployment checklist: 1. Use cv2.imread(path, cv2.IMREAD_UNCHANGED) -- NOT IMREAD_COLOR or IMREAD_GRAYSCALE 2. Use bridge.cv2_to_imgmsg(depth, encoding='16UC1') 3. Convert to meters: depth_m = depth_mm.astype(np.float32) / 1000.0 4. Zero values mean 'no depth' (background) -- handled by depth_image_proc 5. depth_scale=1.0 in BOP metadata confirms mm units in PNG ================================================================================