def pose2d_estimation(self):
parser = ArgParse.create_parser()
args, _ = parser.parse_known_args()
args.checkpoint = False
args.unlabeled = self.folder
args.resume = config["resume"]
args.stacks = config["num_stacks"]
args.test_batch = config["batch_size"]
args.img_res = [
config["heatmap_shape"][0] * 4, config["heatmap_shape"][1] * 4
]
args.hm_res = config["heatmap_shape"]
args.num_classes = config["num_predict"]
args.max_img_id = self.state.num_images - 1
# run the main, it will save the heatmaps and predictions in the image folder
_, _ = pose2d_main(args)
# makes sure cameras use the latest heatmaps and predictions
self.set_cameras()
self.set_mode(Mode.POSE)
for ip in self.image_pose_list:
ip.cam = self.camNetAll[ip.cam.cam_id]
for ip in self.image_pose_list_bot:
ip.cam = self.camNetAll[ip.cam.cam_id]
def pose2d_estimation(self):
from deepfly.pose2d import ArgParse
from deepfly.pose2d.drosophila import main
parser = ArgParse.create_parser()
args, _ = parser.parse_known_args()
args.checkpoint = False
args.unlabeled = self.folder
args.resume = config["resume"]
args.stacks = config["num_stacks"]
args.test_batch = config["batch_size"]
args.img_res = [
config["heatmap_shape"][0] * 4, config["heatmap_shape"][1] * 4
]
args.hm_res = config["heatmap_shape"]
args.num_classes = config["num_predict"]
args.max_img_id = self.state.num_images - 1
# run the main, it will save the heatmaps and predictions in the image folder
_, _ = main(args)
# makes sure cameras use the latest heatmaps and predictions
self.set_cameras()
self.camNet = self.camNetLeft
self.set_mode(Mode.POSE)
self.set_view(self.state.view)
self.update_frame()
def pose2d_estimation(self, overwrite=True):
"""Runs the pose2d estimation on self.input_folder.
Parameters:
overwrite: whether to overwrite existing pose estimation results (default: True)
"""
parser = ArgParse.create_parser()
args, _ = parser.parse_known_args()
args.checkpoint = False
args.unlabeled = self.input_folder
args.output_folder = self.output_subfolder
args.resume = config["resume"]
args.stacks = config["num_stacks"]
args.test_batch = config["batch_size"]
args.img_res = [
config["heatmap_shape"][0] * 4, config["heatmap_shape"][1] * 4
]
args.hm_res = config["heatmap_shape"]
args.num_classes = config["num_predict"]
args.max_img_id = self.max_img_id
args.overwrite = overwrite
pose2d_main(args) # will write output files in output directory
self.set_cameras(
) # makes sure cameras use the latest heatmaps and predictions
def parse_cli_args():
parser = argparse.ArgumentParser(
description = "DeepFly3D pose estimation"
)
parser.add_argument(
"input_folder",
help="Folder containing unlabeled images."
)
parser.add_argument(
"-n", "--num-images-max",
help="Maximal number of images to process.",
default=math.inf,
type=int
)
parser = ArgParse.add_arguments(parser)
return parser.parse_args()
def pose2d_estimation(self):
# print('begain pose2d estimation','\n')
parser = ArgParse.create_parser()
args, _ = parser.parse_known_args()
# print(args, '\n pose2d_estimation')
args.checkpoint = False
args.unlabeled = self.folder
# args.resume = config["resume"]
# print(args.resume,'\nresume\n')
args.resume = '../../weights/sh8_mpii.tar'
args.stacks = config["num_stacks"]
args.test_batch = config["batch_size"]
args.img_res = [config["heatmap_shape"][0] * 4, config["heatmap_shape"][1] * 4]
args.hm_res = config["heatmap_shape"]
args.num_classes = config["num_predict"]
args.max_img_id = self.state.num_images - 1
print('self.state.num_images', self.state.num_images)
# run the main, it will save the heatmaps and predictions in the image folder
args.train_joints = np.arange(args.num_classes)
acc_joints_tmp = []
for i in range(3):
p = [(i * 5) + k for k in args.acc_joints]
acc_joints_tmp.extend(p)
args.acc_joints = acc_joints_tmp
# print(args.acc_joints, '^'*100) #TODO
args.num_images_max = 2
args.workers = 2
# args.checkpoint = os.path.join(args.checkpoint, get_time())
# ,
# "_{}_{}_{}_{}_{}_{}_{}".format(
# "predict" if args.unlabeled else "training",
# args.arch,
# args.stacks,
# args.img_res,
# args.blocks,
# "mv" if args.multiview else "",
# "temp" if args.multiview else "",
# args.name,
# ),
# )
# args.checkpoint = (
# args.checkpoint.replace(" ", "_").replace("(", "_").replace(")", "_")
# )
# args.checkpoint = args.checkpoint.replace("__", "_").replace("--", "-")
# # getLogger('df3d').debug("Checkpoint dir: {}".format(args.checkpoint))
# args.checkpoint_image_dir = os.path.join(args.checkpoint, "./images/")
# create checkpoint dir and image dir
# if args.unlabeled is None:
# if not isdir(args.checkpoint):
# mkdir_p(args.checkpoint)
# if not isdir(args.checkpoint_image_dir):
# mkdir_p(args.checkpoint_image_dir)
# if args.carry and not isdir(
# os.path.join(args.annotation_path, args.unlabeled + "_network")
# ):
# mkdir_p(os.path.join(args.annotation_path, args.unlabeled + "_network"))
# if args.carry and not isdir(
# os.path.join(args.data_folder, args.unlabeled + "_network")
# ):
# mkdir_p(os.path.join(args.data_folder, args.unlabeled + "_network"))
# print(args, '\npose2d_estimation(self)')
_, _ = pose2d_main(args)
# print('pose2d_main(args)', '$' * 70)
# makes sure cameras use the latest heatmaps and predictions
self.set_cameras()
self.set_mode(Mode.POSE)
# print('self.set_mode(Mode.POSE)', '$'*70)
for ip in self.image_pose_list:
ip.cam = self.camNetAll[ip.cam.cam_id]
for ip in self.image_pose_list_bot:
ip.cam = self.camNetAll[ip.cam.cam_id]
self.update_frame()