trt wpodnet lp detection inference file

trt_run_inference.py

import numpy as np
from PIL import Image
import cv2
from src.keras_utils import reconstruct
import pycuda.driver as cuda
import pycuda.autoinit # For automatic creation and cleanup of CUDA context

import tensorrt as trt

import utils

ENGINE_PATH = '/home/kingashar/YOLOV3-LP/new.engine'
CLASSES = ['LP']
INPUT_DATA_TYPE = np.float32
LP_DIMENSIONS = (256, 96)
threshold = 0.95

def im2single(I):
	assert(I.dtype == 'uint8')
	return I.astype('float32')/255.

trt_logger = trt.Logger(trt.Logger.INFO)
runtime = trt.Runtime(trt_logger)
with open(ENGINE_PATH, "rb") as f:
    engine = runtime.deserialize_cuda_engine(f.read())

context = engine.create_execution_context()

stream = cuda.Stream()

host_in = cuda.pagelocked_empty(trt.volume(engine.get_binding_shape(0)), dtype=INPUT_DATA_TYPE)
host_out = cuda.pagelocked_empty(trt.volume(engine.get_binding_shape(1)), dtype=INPUT_DATA_TYPE)
devide_in = cuda.mem_alloc(host_in.nbytes)
devide_out = cuda.mem_alloc(host_out.nbytes)



INPUT_IMAGE_PATH = '/home/kingashar/Downloads/6205_0.jpg' # ADJUST

image = cv2.imread(INPUT_IMAGE_PATH)
image_tensor = im2single(image)
image_tensor = cv2.resize(image,(600,845))
image_tensor = image_tensor.copy()
image_tensor = image_tensor.reshape((1,image_tensor.shape[1],image_tensor.shape[0],image_tensor.shape[2]))
print(image_tensor.shape)
# image_tensor = image_tensor.ravel()


# %matplotlib inline
# import matplotlib.pyplot as plt
# _=plt.imshow(img.transpose(1, 2, 0).astype(np.float32)/255.)
import time

start = time.time()
def infer(img):
    bindings = [int(devide_in), int(devide_out)]
    np.copyto(host_in, img.ravel())
    cuda.memcpy_htod_async(devide_in, host_in, stream)
    context.execute_async(bindings=bindings, stream_handle=stream.handle)
    cuda.memcpy_dtoh_async(host_out, devide_out, stream)
    stream.synchronize()
    return host_out

out = infer(image_tensor)
Yr = np.squeeze(out)
print('Input : {}'.format(INPUT_IMAGE_PATH))
print('Output: {}'.format(out))
print('Prediction: {}'.format(np.squeeze(out)))
print((time.time()-start)*1000)


with open('/home/kingashar/YOLOV3-LP/yr.txt', 'w') as f:
    f.write('{}'.format(str(Yr)))

Thanks for sharing your work! I am using this script and the one for keras_inference both of them give different outputs. If you can throw some light on how to get correct output inTRT inference it would be great and very helpful.

Thanks!