File size: 2,962 Bytes
ae29340 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 | import warnings
warnings.simplefilter("ignore", FutureWarning)
warnings.simplefilter("ignore", UserWarning)
import torch
from torch.profiler import profile, ProfilerActivity
from model.model import Thundernet
from models_repo.model_attention import Thundernet as Thundernet_attention
from models_repo.model_attention_2 import Thundernet as Thundernet_attention2
from models_repo.model_ppm_factors import Thundernet as Thundernet_ppm
import time
import cv2
import numpy as np
import tensorflow as tf
device: torch.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Define input shape
input_shape = (480, 640, 3)
def execute_profiler(model: Thundernet) -> None:
"""
Function to measure de CPU and CUDA times.
It prints the results on the console
Args:
- model: loaded model to profile
Returns:
- None
"""
image = torch.randn(1, 480, 640, 3).cpu().numpy()
# model = torch.jit.trace(model, (image, depth))
with profile(
activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
record_shapes=True,
profile_memory=True,
) as prof:
with torch.no_grad():
_ = model.predict(image)
print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=20))
return
def measure_inference_time(model: Thundernet) -> None:
"""
Function to measure the average inference time
and the FPS of a given AsymFormer model.
It prints the results on the console
Args:
- model: loaded model to profile
Returns:
- None
"""
image = torch.randn(1, 480, 640, 3).cpu().numpy()
for _ in range(5):
_ = model.predict(image)
times = []
for _ in range(20):
tf.constant(0).numpy()
start = time.time()
_ = model.predict(image)
tf.constant(0).numpy()
times.append((time.time() - start) * 1000)
avg_time = sum(times) / len(times)
print(f"Average inference time: {avg_time:.2f} ms")
fps = 1000 / avg_time
print(f"FPS: {fps:.2f}")
return
def main() -> None:
# PATH TO THE BEST MODEL SO FAR (.hdf5)
# weights_path = "D:/RealTimeSemanticSegmentation_Sofia/keras.hdf5"
# weights_path = "C:/Users/user/Documents/Thundernet/pruebas_modelos/32_ppm/BS4_lossBCE_weights_lr_0.00013713842558297858_reg-1.1743577101671763e-05-ep-13-val_loss0.11463435739278793-train_loss0.053004469722509384-val_iou0.8959722518920898-train_iou0.9606077075004578.hdf5"
weights_path = "keras.hdf5"
# Load the model. Change it depending on where it was trained
# ThunderNet = Thundernet_ppm(input_shape=input_shape, resnet_trainable=False, n_classes = 2)
ThunderNet = Thundernet(
input_shape=input_shape, resnet_trainable=False, n_classes=2
)
model = ThunderNet.model
ThunderNet.model.load_weights(weights_path)
execute_profiler(model)
measure_inference_time(model)
return
if __name__ == "__main__":
main()
|