用PaddlePaddle 实现目标检测任务——Paddle Fluid v1.1深度测评

可以看一下这个reader的核心代码：

def reader():

if mode == 'train' and shuffle:

np.random.shuffle(images)

batch_out = []

for image in images:

image_name = image['file_name']

image_path = os.path.join(settings.data_dir, image_name)

im = Image.open(image_path)

if im.mode == 'L':

im = im.convert('RGB')

im_width, im_height = im.size

im_id = image['id']

# layout: category_id | xmin | ymin | xmax | ymax | iscrowd

bbox_labels = []

annIds = coco.getAnnIds(imgIds=image['id'])

anns = coco.loadAnns(annIds)

for ann in anns:

bbox_sample = []

# start from 1, leave 0 to background

bbox_sample.append(float(ann['category_id']))

bbox = ann['bbox']

xmin, ymin, w, h = bbox

xmax = xmin + w

ymax = ymin + h

bbox_sample.append(float(xmin) / im_width)

bbox_sample.append(float(ymin) / im_height)

bbox_sample.append(float(xmax) / im_width)

bbox_sample.append(float(ymax) / im_height)

bbox_sample.append(float(ann['iscrowd']))

bbox_labels.append(bbox_sample)

im, sample_labels = preprocess(im, bbox_labels, mode, settings)

sample_labels = np.array(sample_labels)

if len(sample_labels) == 0: continue

im = im.astype('float32')

boxes = sample_labels[:, 1:5]

lbls = sample_labels[:, 0].astype('int32')

iscrowd = sample_labels[:, -1].astype('int32')

if 'cocoMAP' in settings.ap_version:

batch_out.append((im, boxes, lbls, iscrowd,

[im_id, im_width, im_height]))

else:

batch_out.append((im, boxes, lbls, iscrowd))

if len(batch_out) == batch_size:

yield batch_out

batch_out = []

可以看到，这里的reader是一个生成器，逐个batch把数据load进内存。在数据读取过程中，需要注意一下几点：

1. 数据集需要放在项目的data目录下，reader通过annotations下的instances_train2017.json文件区分训练集和验证集，不需要在data目录下用文件夹区分训练集和验证集。

（编辑：好传媒门户网）

【声明】本站内容均来自网络，其相关言论仅代表作者个人观点，不代表本站立场。若无意侵犯到您的权利，请及时与联系站长删除相关内容!