Created by: guru4elephant

To make image processing easy, I propose to use torchvision style API for user to do image preprocessing. Example code is as follows:

from paddle_serving_app.reader import File2Image, Sequential, Normalize, CenterCrop, Resize

seq = Sequential([
    File2Image(), CenterCrop(30),
    Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Resize((5, 5))
])

url = "daisy.jpg"
for x in range(100):
    img = seq(url)
    print(img.shape)

Resnet50 preprocessing for inference can be as follows:

import sys
from paddle_serving_client import Client
from paddle_serving_app.reader import Sequential, File2Image, Resize, CenterCrop, RGB2BGR, Transpose, Div, Normalize
import time

client = Client()
client.load_client_config(sys.argv[1])
client.connect(["127.0.0.1:9393"])

seq = Sequential([
    File2Image(), Resize(256), CenterCrop(224), RGB2BGR(), Transpose((2, 0, 1)),
    Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
print(seq)

start = time.time()
image_file = "daisy.jpg"
for i in range(1000):
    img = seq(image_file)
    fetch_map = client.predict(feed={"image": img}, fetch=["score"])
end = time.time()
print(end - start)

Image Detection with Faster RCNN example code is as follows:

from paddle_serving_client import Client
import sys
from paddle_serving_app.reader import File2Image, Sequential, Normalize, Resize, Transpose, Div, BGR2RGB, RCNNPostprocess
import numpy as np

preprocess = Sequential([
    File2Image(), BGR2RGB(), Div(255.0),
    Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),
    Resize(640, 640), Transpose((2, 0, 1))
])

postprocess = RCNNPostprocess("label_list.txt", "output")

client = Client()
client.load_client_config(sys.argv[1])
client.connect(['127.0.0.1:9393'])

for i in range(100):
    im = preprocess(sys.argv[2])
    fetch_map = client.predict(
        feed={
            "image": im,
            "im_info": np.array(list(im.shape[1:]) + [1.0]),
            "im_shape": np.array(list(im.shape[1:]) + [1.0])
        },
        fetch=["multiclass_nms"])
    fetch_map["image"] = sys.argv[2]
    postprocess(fetch_map)

Image Segmentation Example

from paddle_serving_client import Client
from paddle_serving_app.reader import Sequential, File2Image, Resize, Transpose, BGR2RGB, SegPostprocess
import sys
import cv2

client = Client()
client.load_client_config(sys.argv[1])
client.connect(["127.0.0.1:9494"])

preprocess = Sequential(
    [File2Image(), Resize(
        (512, 512), interpolation=cv2.INTER_LINEAR)])

postprocess = SegPostprocess(2)

for i in range(100):
    filename = sys.argv[2]
    im = preprocess(filename)
    fetch_map = client.predict(feed={"image": im}, fetch=["output"])
    fetch_map["filename"] = filename
    postprocess(fetch_map)