Tell someone a solution to the problem of launching a YOLOv3 model through a TensorRT. I converted weights to tensorrt:
[i]import cv2
import time
import numpy as np
import tensorflow as tf
import tensorflow.contrib.tensorrt as trt
from tensorflow.python.platform import gfile
from PIL import Image
from YOLOv3 import utils
print(‘Done. All lib loaded’)
function to read a “.pb” model
(can be used to read frozen model or TensorRT model)
def read_pb_return_tensors(graph, pb_file, return_elements):
with tf.gfile.FastGFile(pb_file, 'rb') as f:
frozen_graph_def = tf.GraphDef()
frozen_graph_def.ParseFromString(f.read())
with graph.as_default():
return_elements = tf.import_graph_def(frozen_graph_def,
return_elements=return_elements)
input_tensor, output_tensors = return_elements[0], return_elements[1:]
return input_tensor, output_tensors
def read_pb_graph(model):
with gfile.FastGFile(model,‘rb’) as f:
graph_def = tf.GraphDef()
graph_def.ParseFromString(f.read())
return graph_def
config
SIZE = [416, 416] #input image dimension
video_path = ‘/home/nvidia/Загрузки/Rewind - Pedestrians.mp4’
classes = utils.read_coco_names(‘./YOLOv3/coco.names’)
num_classes = len(classes)
TENSORRT_YOLOv3_MODEL = “./model/TensorRT_YOLOv3_2.pb” # to use the TensorRT optimized model
print(‘Model is load’)
get input-output tensor
input_tensor, output_tensors = read_pb_return_tensors(tf.get_default_graph(), TENSORRT_YOLOv3_MODEL, [“Placeholder:0”, “concat_9:0”, “mul_9:0”])
[/i]
Then I try to run the model in a tensorflow session:
with tf.Session (config = tf.ConfigProto (gpu_options = tf.GPUOptions (per_process_gpu_memory_fraction = 0.5))) as sess:
print (‘Read from video’)
vid = cv2.VideoCapture (video_path)
while true:
return_value, frame = vid.read ()
But the process is killed - the memory is full.
What can I do wrong?
P.S. I took the startup examples in this repository GitHub - ardianumam/Tensorflow-TensorRT: This repository is for my YT video series about optimizing a Tensorflow deep learning model using TensorRT. We demonstrate optimizing LeNet-like model and YOLOv3 model, and get 3.7x and 1.5x faster for the former and the latter, respectively, compared to the original models.