Mesh not loaded correctly (ObjectRecognizer crashes)

[toliver]

As discussed in https://groups.google.com/forum/#!topic/object-recognition-kitchen/ijpfJdG9q8E

The ObjectRecognizer.cpp tries to read a mesh called "original" from the DB, but the default way to generate a mesh (i.e. rosrun object_recognition_reconstruction mesh_object --all --commit) doesn't generate a file with such a name.

I modified it to get a bit further. See the following commit.

shadow-robot@2c149db

Then I get the following output. The file is correctly read from the DB, and parsed using the assimp library (aiImportFile()) but for some reason it crashes in line 176 of ObjectRecognizer.cpp (in my modified version) when trying to access the number of indices of the first face of the mesh (face.mNumIndices). This seems to point to a bug in the mesh generator, or a bug in the assimp library. Any ideas?

rosrun object_recognition_core detection -c detection.object.ros.ork 
[ INFO] [1393263269.953285655]: Initialized ROS. node_name: /object_recognition_server
[ INFO] [1393263270.736406354]: System already initialized. node_name: /object_recognition_server
[ INFO] [1393263271.697297591]: System already initialized. node_name: /object_recognition_server
[ INFO] [1393263271.708924056]: Subscribed to topic:/camera/depth_registered/camera_info with queue size of 1
[ INFO] [1393263271.712172667]: Subscribed to topic:/camera/rgb/camera_info with queue size of 1
[ INFO] [1393263271.715300890]: Subscribed to topic:/camera/depth_registered/image with queue size of 1
[ INFO] [1393263271.717938160]: Subscribed to topic:/camera/rgb/image_color with queue size of 1
126 :: 0.5 , 1 , 0.995004 , /base_link , /head_mount_kinect_rgb_optical_frame
Loading model: b34c1fdbfcc566e265f880e1f001353f for object id: b34c1fdbfcc566e265f880e1f000049bInfo,  T0: Load /tmp/filez5OUHT.stl
Info,  T0: Found a matching importer for this file format
Info,  T0: Import root directory is '/tmp/'
Info,  T0: Entering post processing pipeline
Warn,  T0: RemoveVCProcess: AI_CONFIG_PP_RVC_FLAGS is zero.
Info,  T0: Points: 0, Lines: 0, Triangles: 1, Polygons: 0 (Meshes, X = removed)
Warn,  T0: Mesh 0: Not suitable for vcache optimization
Info,  T0: Cache relevant are 0 meshes (0 faces). Average output ACMR is -nan
Info,  T0: Leaving post processing pipeline
Mesh: 0 mNumVertices: 7899
Mesh: 0 mNumFaces: 2633
after 0
face[0] address: 0x3cc10454ba60120c
Segmentation fault (core dumped)

[vrabaud]

hi, I fixed this in trunk, is there any way you can try using the sources ?

[toliver]

Yes, I could test with the source, but where did you fix it (i.e. what code should I pull)? (I couldn't see any recent commit in the master branch).

[vrabaud]

I fixed your original problem but I also added a mandatory build of Assimp3 (with an updated version), please let me know how that goes.

[toliver]

I've just tested it and it works well. Loads the mesh from DB and recognizes the object.

[awesomebytes]

I can confirm I get detections with tabletop now!
The behaviour is still not exactly what I want but I'll look into it a bit further and see if I can make a pull request with the changes. Also prepare a version that works in simulation with REEM robot so anyone can test in there.

[awesomebytes]

As this was fixed, closing.