Hello, how about testing my own video?
yangzhegithub opened this issue · 2 comments
yangzhegithub commented
Hello, how about testing my own video?
huanglianghua commented
Hi, a code snippet for testing on your own data would be like this:
import _init_paths
import neuron.data as data
from trackers import GlobalTrack
cfg_file = 'configs/qg_rcnn_r50_fpn.py'
ckp_file = 'checkpoints/qg_rcnn_r50_fpn_coco_got10k_lasot.pth'
transforms = data.BasicPairTransforms(train=False)
tracker = GlobalTrack(
cfg_file, ckp_file, transforms,
name_suffix='qg_rcnn_r50_fpn')
# TODO: define your own img_files and init_bbox
# img_files: a list of paths to your sequential images
# init_bbox: a 4-element array containing [x1, y1, x2, y2]
out_bboxes = tracker.track(img_files, init_bbox, visualize=True)
Or, if you are dealing with sequential images, you can do manual init
and update
:
tracker.init(img_0, init_bbox)
for ...
out_bbox = tracker.update(img_i)
yangzhegithub commented
@huanglianghua thank you so much, this is a nice job.