OpenPointCloud/NNCTX/pytorch/combine_train_data.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
为了加速创建trainsets，分两次生成data，需要把这两次的data合并
"""
import os
import numpy as np

# root_dir = '/userhome/NNCTX-main/train_datasets/'
# samples = ['andrew10','david10','sarah10']#,'soldier','longdress'
# nfr_per_samples = [6,6,6]#,9,9

data = np.load("/userhome/NNCTX-main/train_datasets/data_100/data1.npy",allow_pickle=True)
ctxs1 = data[()]['ctxs']
counts1 = data[()]['counts']
ply_paths = data[()]['ply_paths']

data = np.load("/userhome/NNCTX-main/train_datasets/data_100/data.npy",allow_pickle=True)
ctxs2 = data[()]['ctxs']
counts2 = data[()]['counts']
ply_paths += data[()]['ply_paths'] #不重要，可以不去重
print(ply_paths)

all_ctxs = np.concatenate((ctxs1,ctxs2),axis=0)#[all_ctxs ; ctxs];
all_counts = np.concatenate((counts1,counts2),axis=0)#[all_ctxs ; ctxs];
nall = all_ctxs.shape[0]

u_ctxs,uinvs = np.unique(all_ctxs, return_inverse=1,axis=0)
nu_ctxs = u_ctxs.shape[0]
u_counts = np.zeros((nu_ctxs,2),'int')

for i_actx in range(nall):
    iuctx = uinvs[i_actx]
    u_counts[iuctx,:] = u_counts[iuctx,:]+all_counts[i_actx,:]
    if i_actx%100000==0:
        print(str(i_actx) +'/' + str(nall ))    

all_ctxs = np.copy(u_ctxs)
all_counts = np.copy(u_counts)
np.save('/userhome/NNCTX-main/train_datasets/data_100/data_all.npy',{'ctxs':all_ctxs,'counts':all_counts,'ply_paths':ply_paths}) #保存一次会覆盖之前的结果，all_ctxs本身包含了之前的结果