Ms-coco hard negs on gypsum

README_HARDNEG_MSCOCO.md

Multi-Class Detection with RetinaNet

Location

On gypsum:

/mnt/nfs/scratch1/arunirc/data/MS-COCO_cls-1_hard-neg

Data format

This shows how to parse the data:

import os.path as osp

ROOT_PATH= '/mnt/nfs/scratch1/arunirc/data/MS-COCO_cls-1_hard-neg'
CLS_ID = 1 # for Person category (fixed)

VID_ID = 101 # have to loop thru these

vid_folder_name = 'hardNeg_video-%d_cls-1_alpha=10_beta=100_validScore=0.60' % VID_ID

vid_frames_folder = osp.join(ROOT_PATH, 'output', vid_folder_name, '101_cls-1')


# annotations file (FDDB-style format) <-- should be quickest to use
txt_annot_file = osp.join(vid_frames_folder, 101_cls-1.txt)

# annotations file (JSON-style format ... *not* exactly MS-COCO)
json_annot_file = osp.join(vid_frames_folder, '101_cls-1_detections.json')

# read in a frame image
img_name = '101_cls-1/101_cls-1_103.jpg'
img_file = osp.join(vid_frames_folder, img_name)

Example for video101:

<DATA_ROOT>/output/hardNeg_video-101_cls-1_alpha=10_beta=100_validScore=0.60/
    |
    |-- vis_101_cls-1
    |
    |-- 101_cls-1
            |
            |-- 101_cls-1.txt
            |-- 101_cls-1_detections.json
            |-- 101_cls-1
                    |--101_cls-1_103.jpg
                    |--101_cls-1_1274.jpg
                    |--101_cls-1_1277.jpg
                    |--101_cls-1_3471.jpg