Spaces:

sohojoe
/

soho-clip-embeddings-explorer

Running

App Files Files Community

soho-clip-embeddings-explorer / api_test.py

sohojoe

use a better sample image

91e4bde about 1 year ago

raw history blame contribute delete

No virus

2.74 kB

	from gradio_client import Client
	import time
	import numpy as np

	import torch

	from api_helper import preprocess_image, encode_numpy_array
	clip_image_size = 224
	num_steps = 1000
	test_image_url = "https://static.wixstatic.com/media/4d6b49_42b9435ce1104008b1b5f7a3c9bfcd69~mv2.jpg/v1/fill/w_454,h_333,fp_0.50_0.50,q_90/4d6b49_42b9435ce1104008b1b5f7a3c9bfcd69~mv2.jpg"


	client = Client("http://127.0.0.1:7860/")

	print("do we have cuda", torch.cuda.is_available())

	def test_text():
	result = client.predict(
	"Howdy!", # str representing string value in 'Input' Textbox component
	api_name="/text_to_embeddings"
	)
	return(result)

	def test_image():
	result = client.predict(
	test_image_url, # str representing filepath or URL to image in 'Image Prompt' Image component
	api_name="/image_to_embeddings"
	)
	return(result)

	def test_image_as_payload(payload):
	result = client.predict(
	payload, # image as string payload
	api_name="/image_as_payload_to_embeddings"
	)
	return(result)

	# performance test for text
	start = time.time()
	for i in range(num_steps):
	test_text()
	end = time.time()
	average_time_seconds = (end - start) / num_steps
	print("Average time for text: ", average_time_seconds, "s")
	print("Average time for text: ", average_time_seconds * 1000, "ms")
	print("Number of predictions per second for text: ", 1 / average_time_seconds)

	# performance test for image
	start = time.time()
	for i in range(num_steps):
	test_image()
	end = time.time()
	average_time_seconds = (end - start) / num_steps
	print("Average time for image: ", average_time_seconds, "s")
	print("Average time for image: ", average_time_seconds * 1000, "ms")
	print("Number of predictions per second for image: ", 1 / average_time_seconds)



	# download image from url
	import requests
	from PIL import Image
	from io import BytesIO
	response = requests.get(test_image_url)
	input_image = Image.open(BytesIO(response.content))
	input_image = input_image.convert('RGB')
	# convert image to numpy array
	input_image = np.array(input_image)

	if input_image.shape[0] > clip_image_size or input_image.shape[1] > clip_image_size:
	input_image = preprocess_image(input_image, clip_image_size)
	payload = encode_numpy_array(input_image)

	# performance test for image as payload
	start = time.time()
	for i in range(num_steps):
	test_image_as_payload(payload)
	end = time.time()
	average_time_seconds = (end - start) / num_steps
	print("Average time for image as payload: ", average_time_seconds, "s")
	print("Average time for image as payload: ", average_time_seconds * 1000, "ms")
	print("Number of predictions per second for image as payload: ", 1 / average_time_seconds)