selected_coco_cls_dets.md

Description of selected_coco_cls_dets.hdf5 file

HDF5 Directory Structure

.
+-- global_id1
|   +-- boxes_scores_rpn_ids
|   +-- start_end_ids
+-- global_id2
|   +-- boxes_scores_rpn_ids
|   +-- start_end_ids
...

Each global_id is an hdf5 group with boxes_scores_rpn_ids and start_end_ids as hdf5 datasets.

HDF5 datasets description

• boxes_scores_rpn_ids is a Nx6 (=4+1+1) dimensional numpy array with each row containing the box coordinates ([x1,y1,x2,y2] where (x1,y1) and (x2,y2) are the top-left and bottom-right coordinates respectively), score for the selected class, and index of the box in the list of 300 boxes proposed by RPN in the Faster-RCNN framework.

• start_end_ids is a 81x2 dimensional numpy array with i^th row containing the start and end row numbers in box_scores_rpn_ids for i^th class in the list of COCO_CLASSES (see exp/detect_coco_objects/coco_classes.py). So detections for i^th category in COCO_CLASSES for a given global_id are obtained by

import h5py
from data.coco_classes import COCO_CLASSES

f = h5py.File(selected_coco_cls_dets_hdf5_path,'r')
cls_name = COCO_CLASSES[i]
start_id, end_id = f[global_id]['start_end_ids'][i]
dets = f[global_id]['boxes_scores_rpn_ids'][start_id:end_id]
boxes = dets[:,:4] # Box coordinates
scores = dets[:,4] # Scores for object category cls_name
rpn_ids = dets[:,5] # ID of the box in the list of predictions made by faster-rcnn (an integer in [0,300))