Caltech Pedestrian¶
The Caltech Pedestrian Dataset consists of approximately 10 hours of 640x480 30Hz video taken from a vehicle driving through regular traffic in an urban environment. About 250,000 frames (in 137 approximately minute long segments) with a total of 350,000 bounding boxes and 2300 unique pedestrians were annotated.
The annotation includes temporal correspondence between bounding boxes and detailed occlusion labels.
Use cases¶
Pedestrian detection in images/videos.
Properties¶
name
: caltech_pedestriankeywords
: image_processing, detection, pedestriandataset size
: 11,9 GBis downloadable
: yes
Note
The detection
tasks contains 1/30 of all frames of each video.
The detection_10x
tasks contains 1/3 of all frames of each video.
The detection_30x
tasks has all the frames of each video.
Tasks ending with _clean
have bounding boxes with small area (less than 5px width/height) discarded.
These are mostly due to bad annotations and are kept from these tasks.
Tasks¶
detection (default)¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped = dbc.load('caltech_pedestrian')
>>> caltech_ped
DataLoader: "caltech_pedestrian" (detection task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos.sets
: train, testmetadata file size in disk
: 524,0 kBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(6365,4)
│ ├── boxesv # dtype=np.float, shape=(6365,4)
│ ├── classes # dtype=np.uint8, shape=(6365,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(6365,)
│ ├── image_filenames # dtype=np.uint8, shape=(6365,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(4250,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(6365,6)
│ ├── occlusion # dtype=np.float, shape=(6365,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(2223,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(2223,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,2033)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(2223,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,5081)
│
└── test/
├── boxes # dtype=np.float, shape=(5142,4)
├── boxesv # dtype=np.float, shape=(5142,4)
├── classes # dtype=np.uint8, shape=(5142,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(5142,)
├── image_filenames # dtype=np.uint8, shape=(5142,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(4024,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(5142,6)
├── occlusion # dtype=np.float, shape=(5142,)
├── list_boxes_per_image # dtype=np.int32, shape=(2152,13)
├── list_boxesv_per_image # dtype=np.int32, shape=(2152,13)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,2014)
├── list_object_ids_per_image # dtype=np.int32, shape=(2152,13)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,4401)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
detection_clean¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped_clean = dbc.load('caltech_pedestrian', 'detection_clean')
>>> caltech_ped_clean
DataLoader: "caltech_pedestrian" (detection_clean task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos. Very small annotations (<5px height/width) have been discarded.sets
: train, testmetadata file size in disk
: 728,4 kBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(6313,4)
│ ├── boxesv # dtype=np.float, shape=(6313,4)
│ ├── classes # dtype=np.uint8, shape=(6313,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(6313,)
│ ├── image_filenames # dtype=np.uint8, shape=(6313,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(4250,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(6313,6)
│ ├── occlusion # dtype=np.float, shape=(6313,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(2218,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(2218,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,2027)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(2218,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,5033)
│
└── test/
├── boxes # dtype=np.float, shape=(5109,4)
├── boxesv # dtype=np.float, shape=(5109,4)
├── classes # dtype=np.uint8, shape=(5109,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(5109,)
├── image_filenames # dtype=np.uint8, shape=(5109,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(4024,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(5109,6)
├── occlusion # dtype=np.float, shape=(5109,)
├── list_boxes_per_image # dtype=np.int32, shape=(2148,13)
├── list_boxesv_per_image # dtype=np.int32, shape=(2148,13)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,2010)
├── list_object_ids_per_image # dtype=np.int32, shape=(2148,13)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,4371)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
detection_10x¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped_10x = dbc.load('caltech_pedestrian', 'detection_10x')
>>> caltech_ped_10x
DataLoader: "caltech_pedestrian" (detection_10x task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos. It contains 10x more annotations than the default task (‘detection’).sets
: train, testmetadata file size in disk
: 4,3 MBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(64063,4)
│ ├── boxesv # dtype=np.float, shape=(64063,4)
│ ├── classes # dtype=np.uint8, shape=(64063,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(64063,)
│ ├── image_filenames # dtype=np.uint8, shape=(64063,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(42782,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(64063,6)
│ ├── occlusion # dtype=np.float, shape=(64063,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(22356,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(22356,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,20480)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(22356,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,51092)
│
└── test/
├── boxes # dtype=np.float, shape=(51451,4)
├── boxesv # dtype=np.float, shape=(51451,4)
├── classes # dtype=np.uint8, shape=(51451,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(51451,)
├── image_filenames # dtype=np.uint8, shape=(51451,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(40465,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(51451,6)
├── occlusion # dtype=np.float, shape=(51451,)
├── list_boxes_per_image # dtype=np.int32, shape=(21653,14)
├── list_boxesv_per_image # dtype=np.int32, shape=(21653,14)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,20239)
├── list_object_ids_per_image # dtype=np.int32, shape=(21653,14)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,44095)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
detection_10x_clean¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped_10x_clean = dbc.load('caltech_pedestrian', 'detection_10x_clean')
>>> caltech_ped_10x_clean
DataLoader: "caltech_pedestrian" (detection_10x_clean task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos. It contains 10x more annotations than the default task (‘detection’). Very small annotations (<5px height/width) have been discarded.sets
: train, testmetadata file size in disk
: 4,3 MBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(63538,4)
│ ├── boxesv # dtype=np.float, shape=(63538,4)
│ ├── classes # dtype=np.uint8, shape=(63538,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(63538,)
│ ├── image_filenames # dtype=np.uint8, shape=(63538,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(42782,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(63538,6)
│ ├── occlusion # dtype=np.float, shape=(63538,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(22303,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(22303,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,20422)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(22303,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,50605)
│
└── test/
├── boxes # dtype=np.float, shape=(51079,4)
├── boxesv # dtype=np.float, shape=(51079,4)
├── classes # dtype=np.uint8, shape=(51079,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(51079,)
├── image_filenames # dtype=np.uint8, shape=(51079,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(40465,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(51079,6)
├── occlusion # dtype=np.float, shape=(51079,)
├── list_boxes_per_image # dtype=np.int32, shape=(21590,14)
├── list_boxesv_per_image # dtype=np.int32, shape=(21590,14)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,20173)
├── list_object_ids_per_image # dtype=np.int32, shape=(21590,14)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,43748)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
detection_30x¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped_30x = dbc.load('caltech_pedestrian', 'detection_30x')
>>> caltech_ped_30x
DataLoader: "caltech_pedestrian" (detection_30x task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos. It contains 10x more annotations than the default task (‘detection’).sets
: train, testmetadata file size in disk
: 12,0 MBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(192185,4)
│ ├── boxesv # dtype=np.float, shape=(192185,4)
│ ├── classes # dtype=np.uint8, shape=(192185,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(192185,)
│ ├── image_filenames # dtype=np.uint8, shape=(192185,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(128419,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(192185,6)
│ ├── occlusion # dtype=np.float, shape=(192185,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(67083,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(67083,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,61439)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(67083,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,153234)
│
└── test/
├── boxes # dtype=np.float, shape=(154436,4)
├── boxesv # dtype=np.float, shape=(154436,4)
├── classes # dtype=np.uint8, shape=(154436,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(154436,)
├── image_filenames # dtype=np.uint8, shape=(154436,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(121465,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(154436,6)
├── occlusion # dtype=np.float, shape=(154436,)
├── list_boxes_per_image # dtype=np.int32, shape=(64999,14)
├── list_boxesv_per_image # dtype=np.int32, shape=(64999,14)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,60748)
├── list_object_ids_per_image # dtype=np.int32, shape=(64999,14)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,132324)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
detection_30x_clean¶
How to use¶
>>> # import the package
>>> import dbcollection as dbc
>>>
>>> # load the dataset
>>> caltech_ped_30x_clean = dbc.load('caltech_pedestrian', 'detection_30x_clean')
>>> caltech_ped_30x_clean
DataLoader: "caltech_pedestrian" (detection_30x_clean task)
Properties¶
primary use
: object detectiondescription
: Contains image filenames, classes and bounding box annotations for pedestrian detection in images/videos. It contains 10x more annotations than the default task (‘detection’). Very small annotations (<5px height/width) have been discarded.sets
: train, testmetadata file size in disk
: 11,9 MBhas annotations
: yeswhich
:- labels for each class/category.
- bounding box of pedestrians.
- occlusion % of pedestrian detections.
HDF5 file structure¶
/
├── train/
│ ├── boxes # dtype=np.float, shape=(190598,4)
│ ├── boxesv # dtype=np.float, shape=(190598,4)
│ ├── classes # dtype=np.uint8, shape=(190598,10) (note: string in ASCII format)
│ ├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
│ ├── id # dtype=np.int32, shape=(190598,)
│ ├── image_filenames # dtype=np.uint8, shape=(190598,90) (note: string in ASCII format)
│ ├── image_filenames_unique # dtype=np.uint8, shape=(128419,90) (note: string in ASCII format)
│ ├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
│ ├── object_ids # dtype=np.int32, shape=(190598,6)
│ ├── occlusion # dtype=np.float, shape=(190598,)
│ ├── list_boxes_per_image # dtype=np.int32, shape=(66923,22)
│ ├── list_boxesv_per_image # dtype=np.int32, shape=(66923,22)
│ ├── list_image_filenames_per_class # dtype=np.int32, shape=(4,61274)
│ ├── list_object_ids_per_image # dtype=np.int32, shape=(66923,22)
│ └── list_objects_ids_per_class # dtype=np.int32, shape=(4,151768)
│
└── test/
├── boxes # dtype=np.float, shape=(153305,4)
├── boxesv # dtype=np.float, shape=(153305,4)
├── classes # dtype=np.uint8, shape=(153305,10) (note: string in ASCII format)
├── classes_unique # dtype=np.uint8, shape=(4,10) (note: string in ASCII format)
├── id # dtype=np.int32, shape=(153305,)
├── image_filenames # dtype=np.uint8, shape=(153305,90) (note: string in ASCII format)
├── image_filenames_unique # dtype=np.uint8, shape=(121465,90) (note: string in ASCII format)
├── object_fields # dtype=np.uint8, shape=(6,16) (note: string in ASCII format)
├── object_ids # dtype=np.int32, shape=(153305,6)
├── occlusion # dtype=np.float, shape=(153305,)
├── list_boxes_per_image # dtype=np.int32, shape=(64801,14)
├── list_boxesv_per_image # dtype=np.int32, shape=(64801,14)
├── list_image_filenames_per_class # dtype=np.int32, shape=(4,60537)
├── list_object_ids_per_image # dtype=np.int32, shape=(64801,14)
└── list_objects_ids_per_class # dtype=np.int32, shape=(4,131273)
Fields¶
boxes
: bounding boxesavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
boxesv
: bounding boxes (visible)available in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1note
: bbox format (x1,y1,x2,y2)
classes
: class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
classes
: unique class namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
id
: label idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1
image_filenames
: image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
image_filenames
: unique image file path + namesavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII format
object_fields
: list of field names of the object id listavailable in
: train, testdtype
: np.uint8is padded
: Truefill value
: 0note
: strings stored in ASCII formatnote
: key field (field name aggregator)
object_ids
: list of field idsavailable in
: train, testdtype
: np.int32is padded
: Falsefill value
: -1note
: key field (field id aggregator)
occlusion
: occlusion percentageavailable in
: train, testdtype
: np.floatis padded
: Falsefill value
: -1
list_boxes_per_image
: list of bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_boxesv_per_image
: list of (visible) bounding boxes per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_image_filenames_per_class
: list of image per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_object_ids_per_image
: list of object ids per imageavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list
list_objects_ids_per_class
: list of object ids per classavailable in
: train, testdtype
: np.int32is padded
: Truefill value
: -1note
: pre-ordered list