scenenet.proto

syntax = "proto2";

package scenenet;

message SceneLayout {
    enum LayoutType {
        BATHROOM = 1;
        BEDROOM = 2;
        KITCHEN = 3;
        LIVING_ROOM = 4;
        OFFICE = 5;
    }
    optional LayoutType layout_type = 1;
    // This is the name of the SceneNet model used for the layout
    optional string model = 2;
}

message LightInfo {
    enum LightType {
        SPHERE = 1;
        PARALLELOGRAM = 2;
    }
    optional LightType light_type = 1;
    // Light intensity
    optional Power light_output = 2;
    // This is the center for sphere type lights. And corner for others
    optional Position position = 3;
    // This is only for SPHERE lights
    optional float radius = 4;
    // This is only for PARALLELOGRAM lights
    optional Position v1 = 5;
    optional Position v2 = 6;
}

message RandomObjectInfo {
    optional string shapenet_hash = 1;
    optional float height_meters = 2;
    message Transformation {
        // The 3x4 matrix is as follows:
        // rotation_mat11 rotation_mat12 rotation_mat13 translation_x
        // rotation_mat21 rotation_mat22 rotation_mat23 translation_y
        // rotation_mat31 rotation_mat32 rotation_mat33 translation_y
        optional float translation_x = 1;
        optional float translation_y = 2;
        optional float translation_z = 3;
        optional float rotation_mat11 = 4;
        optional float rotation_mat12 = 5;
        optional float rotation_mat13 = 6;
        optional float rotation_mat21 = 7;
        optional float rotation_mat22 = 8;
        optional float rotation_mat23 = 9;
        optional float rotation_mat31 = 10;
        optional float rotation_mat32 = 11;
        optional float rotation_mat33 = 12;
    }
    // The transformation gives the transformation applies to an object, about
    // the center of the base plane of its axis-aligned bounding box.
    optional Transformation object_pose = 3;
}

message Instance {
    optional int32 instance_id = 1;
    optional string semantic_wordnet_id = 2;
    optional string semantic_english = 3;
    enum InstanceType {
        // This is the instance type when no object is present, e.g. because of
        // looking out a window into nothingness
        BACKGROUND = 1;
        // This is an object that is hard coded into the layout and does not
        // move.  This type does not have a transformation or shapenet hash
        LAYOUT_OBJECT = 2;
        // This is a randomly positioned light source
        LIGHT_OBJECT = 3;
        // This means the object is a randomly positioned shapenet object. The
        // object has a transformation and scale parameter in the object_info
        // variable.
        RANDOM_OBJECT = 4;
    }
    optional InstanceType instance_type = 4;
    // This information is only filled in for the respective type
    optional LightInfo light_info = 5;
    optional RandomObjectInfo object_info = 6;
}

message Power {
    optional float r = 1;
    optional float g = 2;
    optional float b = 3;
}

message Position {
    optional float x = 1;
    optional float y = 2;
    optional float z = 3;
}

message Pose {
    // The position of these two points define the camera view. The y vector is
    // defined as [0,1,0].  For an example of how to calculate the camera view
    // coordinate system, see the python codebase.
    optional Position camera = 1;
    optional Position lookat = 2;
    optional float timestamp = 3;
}

message View {
    // These increment by the number of skip frames, i.e. 0,25,50...7475.
    optional int32 frame_num = 1;
    // The photo is rendered by integrating uniformly sampled 
    // exposures between the following two poses
    optional Pose shutter_open = 2;
    optional Pose shutter_close = 3;
}

message Trajectory {
    optional SceneLayout layout = 1;
    // The first instances[0] is always the 'background' and
    // undefined class when for example looking out windows
    repeated Instance instances = 2;
    // These are ordered sequentially for a trajectory
    repeated View views = 3;
    // This stores the path from the root data directory to the trajectory data
    // folder.  If the trajectories are stored as:
    // /path/i/extracted/{val/train}/0/123/photo/0.jpg 
    // then this path will be '0/123' designating the trajectories folder
    optional string render_path = 4;
}

message Trajectories {
    // This is the root list which stores all of the available trajectories
    repeated Trajectory trajectories = 1;
}