preprocess_joints_new.py

import numpy as np
from hand_cropping import crop_hand_joint
import os
import mediapipe as mp


train_img = np.load('data/train_img.npy')
test_img = np.load('data/test_img.npy')
train_lab = np.load('data/train_lab.npy')
test_lab = np.load('data/test_lab.npy')

hands = mp.solutions.hands.Hands(static_image_mode=True,
                        max_num_hands=1,
                        min_detection_confidence=0.5)
train_joints = []
n_train_lab = []
for img,lab in zip(train_img,train_lab):
	joint = crop_hand_joint(img,hands)
	if joint is None:
		continue
	train_joints.append(joint[0])
	n_train_lab.append(lab)

train_joints = np.concatenate(train_joints,axis=0)
n_train_lab = np.array(n_train_lab)

test_joints = []
n_test_lab = []
for img,lab in zip(test_img,test_lab):
	joint = crop_hand_joint(img,hands)
	if joint is None:
		continue
	test_joints.append(joint[0])
	n_test_lab.append(lab)

test_joints = np.concatenate(test_joints,axis=0)
n_test_lab = np.array(n_test_lab)

np.save('data/train_joints',train_joints)
np.save('data/train_lab_joint',n_train_lab)

np.save('data/test_joints',test_joints)
np.save('data/test_lab_joint',n_test_lab)