Frame pose information for the Kitchen scenes dataset:
======================================================
The pose information for the frames of each scene is in the "scene_pose_info/gmu_scene_00#_reconstruct_info_frame_sort.mat" files than can be loaded in MATLAB.
Each mat file contains a struct 'frames' where each index corresponds to information for a frame:
>> frames(1)
ans =
        camInd: 425
    intrinsics: [1056.3 -0.0014655 0]
          Rw2c: [3x3 double]
          Tw2c: [-3.4797 1.6816 -1.9044]
       imgName: 'rgb_0.png'

 	1) Use the imgName provided to load the correct frame, and ignore the camInd.

 	2) Rw2c means Rotation world to camera, Tw2c is Translation.
 
	3) We followed the instructions from https://www.cs.cornell.edu/~snavely/bundler/bundler-v0.4-manual.html#S6 for the bundler files on how to project a 3D point on a camera frame. k1 and k2 shown in the instructions are the distortion parameters from frames(i).intrinsics(2) and frames(i).intrinsics(3) for frame i. For focal length f, we did not use the one estimated by COLMAP (frames(i).intrinsics(1)) but we used fx,fy,cx,cy found in the calib_color.yaml file (from the calibration of the Kinect) that is included.
======================================================


Scene Reconstructions:
======================
The reconstructions are provided in "scene_reconstructions/gmu_scene_00#_reconstruction.ply" files that can be visualized in Meshlab. The files can be loaded in Matlab to get the 3D point cloud and RGB color information using the code provided here:
https://www.mathworks.com/matlabcentral/fileexchange/5355-toolbox-graph/content/toolbox_graph/read_ply.m
======================


Annotation:
===========
We provide bounding box in image space and 3D annotation for the objects. 
We used the object names provided in the BigBird dataset and labeled them alphabetically.
The bounding box annotations for the 11 BigBird objects are provided in "scene_annotation/bboxes/gmu_scene_00#_annotated_bboxes.mat",
files that can be loaded in Matlab. 
Annotations for all objects are provided in the "scene_annotation/bboxes/gmu_scene_00#_annotated_bboxes_withExtra.mat" files.

Each file contains a struct 'bboxes' shown in the following example:
>> bboxes{frameIndex}(bboxIndex)
ans = 
    category: {'coca_cola_glass_bottle'}
       label: 1
         top: 207
        left: 355
      bottom: 427
       right: 431
     imgName: 'rgb_0.png'

The categories and corresponding labels provided are:
	BigBird Objects:
coca_cola_glass_bottle 1
coffee_mate_french_vanilla 2
honey_bunches_of_oats_honey_roasted 3
hunts_sauce 4
mahatma_rice 5
nature_valley_soft_baked_oatmeal_squares_cinnamon_brown_sugar 6
nature_valley_sweet_and_salty_nut_almond 7
palmolive_orange 8
pop_secret_light_butter 9
pringles_bbq 10
red_bull 11
	Extra Objects (and scenes they appear)
bowl_1 	12  	scenes 1,2
bowl_2 	13  	scenes 1,2
bowl_3	14	scenes 5,6
bowl_4	15	scenes 5,6
bowl_5	16	scene 9
mug_1  	17  	scenes 1,2
mug_2   18  	scenes 1,2
mug_3   19	scenes 5,6
mug_4	20	scenes 5,6
mug_5	21	scene 9
salt    22  	scenes 1,2
cleaner 23      scenes 3,4

The cropped 3D point cloud for each object can be found in "scene_annotation/objects3D/gmu_scene_00#/objectK.ply", where K is the label of the object following the list above. The files can be visualized and loaded in Matlab as explained in the Scene Reconstructions paragraph above. The extra objects are also provided in a separate folder for each scene.

The 3D point labels for each dense scene point cloud can be found in "scene_annotation/point_3D_labeling/scene_#_3D_labels.mat". Each file contains a vector the size of the dense point cloud which holds the label for each 3D point. The labels are the same as shown above plus Background which is 0.
===========