about fps

[1999zsh]

Star RTDETR
请先在RTDETR主页点击star以支持本项目
Star RTDETR to help more people discover this project.

---

Describe the bug
A clear and concise description of what the bug is.
If applicable, add screenshots to help explain your problem.

To Reproduce
Steps to reproduce the behavior.

Dear author.
here is my method of computing fps:

from src.nn.backbone import PResNet
from src.zoo.rtdetr import HybridEncoder, RTDETRTransformer, RTDETR, RTDETRPostProcessor
from PIL import Image
from torchvision import transforms
import cv2
import os
import time

device = "cuda"
backbone = PResNet(depth=18, freeze_norm=False, pretrained=False, return_idx=[1, 2, 3])
encoder = HybridEncoder(in_channels=[128, 256, 512], enc_act='gelu', expansion=0.5, eval_spatial_size=[640, 640])
decoder = RTDETRTransformer(num_classes=10, eval_idx=-1, num_decoder_layers=3, num_denoising=100, 
                            feat_channels=[256, 256, 256], feat_strides=[8, 16, 32],
                            hidden_dim=256, num_levels=3, num_queries=300, eval_spatial_size=[640, 640])
model = RTDETR(backbone=backbone, encoder=encoder, decoder=decoder, 
               multi_scale=[480, 512, 544, 576, 608, 640, 640, 640, 672, 704, 736, 768, 800]).to(device)

postprocessor = RTDETRPostProcessor(num_classes=10, remap_mscoco_category=True)

weights = torch.load("/home/zsh/rtdetr/checkpoint0071.pth")
model.load_state_dict(weights['model'])
model.eval()

font = cv2.FONT_HERSHEY_SIMPLEX

transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Resize(size=(640, 640), antialias=True)
])

xs = []
oris = []
for file in os.listdir("/home/zsh/datasets/rtdetr_visdrone/VisDrone2019-DET-val/images/"):
    x = Image.open(f"/home/zsh/datasets/rtdetr_visdrone/VisDrone2019-DET-val/images/{file}").convert("RGB")
    ori_size = torch.tensor(x.size, dtype=torch.int).to(device)
    x = transform(x)
    x = x.unsqueeze(0).to(device)
    xs.append(x)
    oris.append(ori_size)

b = time.time()
for x, ori in zip(xs, oris):
    y = model(x)
    result = postprocessor(y, ori)[0]
e = time.time()
print(e-b, len(xs), len(xs)/(e-b))```

fps is 40 on rtx4060. Its different from the paper. So, what's wrong with my code, thank you

[lyuwenyu]

---

We test the speed on T4 GPU with TensorRT FP16.

---