论文地址：PointNet网络
代码地址：https://github.com/charlesq34/pointnet
windows10环境配置可参考这篇文章：windows下运行pointnet（全）
网络图示：
一、训练模型
1）下载测试集
数据下载：PartAnnotation（注意该链接中对应的数据集有两部分，因此下载时两部分务必都要下载）
存放位置：…/part_seg/PartAnnotation。
HDF5文件解压放在pointnet-master/part_seg/hdf5_data。
2）训练模型
可参考Pointnet（part_seg）部件分割网络代码解析：train.py（二）文章，里面有代码详解，以及训练模型的获取。
3）测试集数据解析
二、test.py代码详解
1、选择第80个训练模型进行测试
parser = argparse.ArgumentParser()
parser.add_argument('--model_path', default='train_results/trained_models/epoch_80.ckpt', help='Model checkpoint path')
FLAGS = parser.parse_args()
2、setting
parser = argparse.ArgumentParser()
# 自行定义命令参数获取 model_path ，保存的训练模型
parser.add_argument('--model_path', default='train_results/trained_models/epoch_80.ckpt', help='Model checkpoint path')
FLAGS = parser.parse_args()
# DEFAULT SETTINGS
pretrained_model_path = FLAGS.model_path # 获取保存好的模型
hdf5_data_dir = os.path.join(BASE_DIR, './hdf5_data') # 获取h5数据集
ply_data_dir = os.path.join(BASE_DIR, './PartAnnotation') # 导入测试数据集
gpu_to_use = 0
output_dir = os.path.join(BASE_DIR, './test_results')
output_verbose = True # If true, output all color-coded part segmentation obj files
# MAIN SCRIPT
point_num = 3000 # the max number of points in the all testing data shapes
batch_size = 1
test_file_list = os.path.join(BASE_DIR, 'testing_ply_file_list.txt')
""" testing_ply_file_list.txt为从PartAnnotation数据集中采样出的2874个数据，分别包括
点云数据 / 分割数据 / 实例类别编号"""
oid2cpid = json.load(open(os.path.join(hdf5_data_dir, 'overallid_to_catid_partid.json'), 'r'))
"""oid2cpid读取物体零件编号
[["02691156", 1], ["02691156", 2],....]"""
object2setofoid = {} # oid对象集
for idx in range(len(oid2cpid)):
objid, pid = oid2cpid[idx] # oid对象标识符 pid编号
if not objid in object2setofoid.keys():
object2setofoid[objid] = []
object2setofoid[objid].append(idx) # 创建一个字典，将每个物体编号按顺序0-49索引排序
'''{‘02691156’:[0,1,2,3],'02247898':[4,5],....}'''
''' 获取16类物体和编号的文件，并划分到两个列表中'''
all_obj_cat_file = os.path.join(hdf5_data_dir, 'all_object_categories.txt')
fin = open(all_obj_cat_file, 'r')
'''
objcats： ['02691156','14546466'....]
objnames： ['Airplane','Bag'....]
'''
lines = [line.rstrip() for line in fin.readlines()]
objcats = [line.split()[1] for line in lines]
objnames = [line.split()[0] for line in lines]
on2oid = {objcats[i]:i for i in range(len(objcats))} # on2oid为物体编号对应的索引，共16个
fin.close()
color_map_file = os.path.join(hdf5_data_dir, 'part_color_mapping.json')
color_map = json.load(open(color_map_file, 'r')) # 获取颜色
NUM_OBJ_CATS = 16 # 16个物体对象
NUM_PART_CATS = 50 # 分割成50个类别
'''为物体零件进行分类1-50类对应 {'03642806_2':29,'03642806_1':28,.....}'''
cpid2oid = json.load(open(os.path.join(hdf5_data_dir, 'catid_partid_to_overallid.json'), 'r'))
3、load_pts_seg_files方法
def load_pts_seg_files(pts_file, seg_file, catid):
with open(pts_file, 'r') as f:
pts_str = [item.rstrip() for item in f.readlines()]
pts = np.array([np.float32(s.split()) for s in pts_str], dtype=np.float32)
with open(seg_file, 'r') as f:
'''在单独一个物体中以1,2,3讲不同零件进行分类，得出的零件索引[2 2 2 1 1 1 1 1 .....]'''
part_ids = np.array([int(item.rstrip()) for item in f.readlines()], dtype=np.uint8)
'''cpid2oid为每个物体零件对应的0-50类编号，将单个物体零件的分类通过cpid2oid转换为总的50类别'''
seg = np.array([cpid2oid[catid+'_'+str(x)] for x in part_ids])
return pts, seg
4、predict()方法
def predict():
is_training = False
with tf.device('/gpu:'+str(gpu_to_use)):
pointclouds_ph, input_label_ph = placeholder_inputs()
is_training_ph = tf.placeholder(tf.bool, shape=())
# simple model
# pointclouds_ph:(1,3000,3) seg_pred:(1,16)
pred, seg_pred, end_points = model.get_model(pointclouds_ph, input_label_ph, \
cat_num=NUM_OBJ_CATS, part_num=NUM_PART_CATS, is_training=is_training_ph, \
batch_size=batch_size, num_point=point_num, weight_decay=0.0, bn_decay=None)
# Add ops to save and restore all the variables.//添加操作用来保存和重现所有变量
saver = tf.train.Saver()
# Later, launch the model, use the saver to restore variables from disk, and
# do some work with the model.
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
with tf.Session(config=config) as sess:
if not os.path.exists(output_dir):
os.mkdir(output_dir)
flog = open(os.path.join(output_dir, 'log.txt'), 'w')
# Restore variables from disk.
printout(flog, 'Loading model %s' % pretrained_model_path)
saver.restore(sess, pretrained_model_path) # 导入训练好的模型
printout(flog, 'Model restored.')
# Note: the evaluation for the model with BN has to have some statistics
# Using some test datas as the statistics
batch_data = np.zeros([batch_size, point_num, 3]).astype(np.float32) # [1,3000,3]
total_acc = 0.0
total_seen = 0
total_acc_iou = 0.0
# NUM_OBJ_CATS=16
total_per_cat_acc = np.zeros((NUM_OBJ_CATS)).astype(np.float32) # 每一类正确的个数
total_per_cat_iou = np.zeros((NUM_OBJ_CATS)).astype(np.float32) # 每一类的IOU
total_per_cat_seen = np.zeros((NUM_OBJ_CATS)).astype(np.int32) # 每一类测试的总个数
ffiles = open(test_file_list, 'r') # 获取测试用的数据集，并进行预处理，将其划分为3类列表
lines = [line.rstrip() for line in ffiles.readlines()]
pts_files = [line.split()[0] for line in lines] # 获取点云文件路径（有2874个路径）
seg_files = [line.split()[1] for line in lines] # 获取seg文件路径
labels = [line.split()[2] for line in lines] # 获取物体类别编号
ffiles.close()
len_pts_files = len(pts_files)
for shape_idx in range(len_pts_files):
if shape_idx % 100 == 0:
printout(flog, '%d/%d ...' % (shape_idx, len_pts_files))
# on2oid为物体编号对应的索引，共16个，获取当前数据集的编号对应索引，将其转换为独热编码
cur_gt_label = on2oid[labels[shape_idx]]
cur_label_one_hot = np.zeros((1, NUM_OBJ_CATS), dtype=np.float32)
cur_label_one_hot[0, cur_gt_label] = 1
# 根据shape_idx将pts和seg文件读取处理
pts_file_to_load = os.path.join(ply_data_dir, pts_files[shape_idx])
seg_file_to_load = os.path.join(ply_data_dir, seg_files[shape_idx])
# 将各物体编号都统一到1-50类当中，这个操作非常关键！！！
pts, seg = load_pts_seg_files(pts_file_to_load, seg_file_to_load, objcats[cur_gt_label])
ori_point_num = len(seg)
batch_data[0, ...] = pc_augment_to_point_num(pc_normalize(pts), point_num)
# 预测出的label和seg
label_pred_val, seg_pred_res = sess.run([pred, seg_pred], feed_dict={
pointclouds_ph: batch_data,
input_label_ph: cur_label_one_hot,
is_training_ph: is_training,
})
# 将预测出的label得出
label_pred_val = np.argmax(label_pred_val[0, :])
seg_pred_res = seg_pred_res[0, ...] # 进行降维处理
# 将该物体的索引提取出来
# objacts:['02691156','02773838',...]
# object2setofoid:{'02691156':[0,1,2,3],'02773838':[4,5],...}
iou_oids = object2setofoid[objcats[cur_gt_label]]
# 创建一个0-49的数组，剔除12,13,14,15
non_cat_labels = list(set(np.arange(NUM_PART_CATS)).difference(set(iou_oids)))
mini = np.min(seg_pred_res) # 获取预测中的最小值
seg_pred_res[:, non_cat_labels] = mini - 1000 # 将除12，,1,14,15的其他标签都减小
# 比较12,13,14,15这个位置得数，取最大判断为该类
seg_pred_val = np.argmax(seg_pred_res, axis=1)[:ori_point_num]
# 预测的类与正确实际的类作比较，得出seg的正确率
seg_acc = np.mean(seg_pred_val == seg)
# 将分割的正确率进行累加
total_acc += seg_acc
# 测试总的个数
total_seen += 1
total_per_cat_seen[cur_gt_label] += 1
total_per_cat_acc[cur_gt_label] += seg_acc
# 预测类与正确的比较，相等为1，不等为0
mask = np.int32(seg_pred_val == seg)
# 计算IOU = n_intersect/(n_pred + n_seg - n_intersect)
# n_pred:预测的标签的个数
# n_seg：实际的标签个数
# n_intersect判断正确的标签个数
total_iou = 0.0
iou_log = ''
for oid in iou_oids:
n_pred = np.sum(seg_pred_val == oid)
n_gt = np.sum(seg == oid)
n_intersect = np.sum(np.int32(seg == oid) * mask)
n_union = n_pred + n_gt - n_intersect
iou_log += '_' + str(n_pred)+'_'+str(n_gt)+'_'+str(n_intersect)+'_'+str(n_union)+'_'
if n_union == 0:
total_iou += 1
iou_log += '_1\n'
else:
total_iou += n_intersect * 1.0 / n_union
iou_log += '_'+str(n_intersect * 1.0 / n_union)+'\n'
avg_iou = total_iou / len(iou_oids)
total_acc_iou += avg_iou
total_per_cat_iou[cur_gt_label] += avg_iou
# 对预测结果，保存在obj文件中
if output_verbose:
output_color_point_cloud(pts, seg, os.path.join(output_dir, str(shape_idx)+'_gt.obj'))
output_color_point_cloud(pts, seg_pred_val, os.path.join(output_dir, str(shape_idx)+'_pred.obj'))
output_color_point_cloud_red_blue(pts, np.int32(seg == seg_pred_val),
os.path.join(output_dir, str(shape_idx)+'_diff.obj'))
with open(os.path.join(output_dir, str(shape_idx)+'.log'), 'w') as fout:
fout.write('Total Point: %d\n\n' % ori_point_num)
fout.write('Ground Truth: %s\n' % objnames[cur_gt_label])
fout.write('Predict: %s\n\n' % objnames[label_pred_val])
fout.write('Accuracy: %f\n' % seg_acc)
fout.write('IoU: %f\n\n' % avg_iou)
fout.write('IoU details: %s\n' % iou_log)
printout(flog, 'Accuracy: %f' % (total_acc / total_seen))
printout(flog, 'IoU: %f' % (total_acc_iou / total_seen))
for cat_idx in range(NUM_OBJ_CATS):
printout(flog, '\t ' + objcats[cat_idx] + ' Total Number: ' + str(total_per_cat_seen[cat_idx]))
if total_per_cat_seen[cat_idx] > 0:
printout(flog, '\t ' + objcats[cat_idx] + ' Accuracy: ' + \
str(total_per_cat_acc[cat_idx] / total_per_cat_seen[cat_idx]))
printout(flog, '\t ' + objcats[cat_idx] + ' IoU: '+ \
str(total_per_cat_iou[cat_idx] / total_per_cat_seen[cat_idx]))
三、结果查看
在路径…/part_seg/test_results文件中可以查看测试结果
1、生成的是obj文件：
对同一个椅子obj文件，可以通过CloudCompare和meshlab查看,效果如下两张图（meshlab查看时可以查看到颜色信息，CloudCompare只能查看到灰色点云，推荐使用meshlab）
2、路径…/part_seg/test_results/log.txt文件中存放的是测试时的日志数据
1029
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
