Ostrack python3.9+5060显卡-lasot部分序列为例
第0步
了解自己显卡所需要的算力
https://en.wikipedia.org/wiki/CUDA
在这基础上往上滑会看到算力对应能装的cuda版本
查看自己显卡所允许最高的cuda版本
win+r-->输入cmd进入终端-->nvidia-smi
查看cuda所对应的pytorch，也要注意python版本
以上东西详细在这位博主的帖子
在anaconda中安装cuda-pytorch_anaconda安装cuda-CSDN博客
现在anaconda外部，即终端下载cuda13.0.2
第1步 创建环境
conda create -n ostrack python=3.9
conda activate ostrack
第2步 改install.sh
echo "****************** Installing pytorch ******************" conda install pytorch==1.9.0 torchvision==0.10.0 torchaudio==0.9.0 cudatoolkit=10.2 -c pytorch echo "" echo ""
改成
echo "****************** Installing pytorch ******************" pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128 echo "" echo ""
删掉：
echo "****************** Downgrade setuptools ******************" pip install setuptools==59.5.0 echo "" echo ""
echo "****************** Installing visdom ******************" pip install visdom echo "" echo ""
echo "****************** Installing thop tool for FLOPs and Params computing ******************" pip install thop-0.0.31.post2005241907 echo "" echo ""
终端输入
conda install m2-base
bash.install.sh
第3步 改代码
下载（忘记需不需要了）
pip install setuptools==63.2.0
卸载（这个是为了下一步检查）
pip uninstall -y tikzplotlib
检查
python -c "import torch; import torchvision; import torchaudio; import yaml; import easydict; import cv2; import pandas; import tqdm; import pycocotools; import jpeg4py; import tensorboard; import thop; import lmdb; import scipy; import tensorboardX; import wandb; import timm; print('✅ 所有核心成功！'); print('🔥 Torch 版本:', torch.__version__); print('🔥 GPU 可用:', torch.cuda.is_available()); print('🔥 显卡:', torch.cuda.get_device_name(0))"
正常应输出
✅ 所有核心库安装成功！
🔥 Torch 版本: 2.8.0+cu128
🔥 GPU 可用: True
🔥 显卡: NVIDIA GeForce RTX 5060 Laptop GPU (ostrack02) PS D:\Program\test11\OSTrack-main>
改代码
打开OSTrack-main\lib\train\data\loader.py
注释from torch._six import string_classes
在同一位置加入string_classes = (str, bytes)
打开OSTrack-main\lib\test\evaluation\local.py
from test.evaluation.environment import EnvSettings
替换成
from lib.test.evaluation.environment import EnvSettings
打开OSTrack-main\lib\vis\visdom_cus.py
注释import visdom import visdom.server这俩行
这里省略把权重放到相应位置、数据集放入，修改local文件和数据集文件
输入命令
python tracking/test.py ostrack vitb_384_mae_ce_32x4_ep300 --dataset lasot --threads 4 --num_gpus 1
就能跑了
第4步 生成跟踪结果视频+画框
把其中一个序列的跟踪结果画成视频 + 画框，生成 MP4
将下面代码写入一个py文件，放在根目录下面，直接运行
import os
import cv2
import numpy as np
# ====================== 已经帮你全部改正确 ======================
DATASET_ROOT = "D:/Program/test11/OSTrack-main/data/lasot/electricfan"
RESULT_ROOT = "D:/Program/test11/OSTrack-main/output/test/tracking_results/ostrack/vitb_384_mae_ce_32x4_ep300/lasot"
SEQUENCE_NAME = "electricfan-1"
SAVE_VIDEO_PATH = f"{SEQUENCE_NAME}_tracking_result.mp4"
# =================================================================
# 读取跟踪结果 txt
def load_results(txt_path):
with open(txt_path, 'r') as f:
lines = f.readlines()
bboxes = []
for line in lines:
line = line.strip().replace(',', ' ')
coords = list(map(float, line.split()))
x1, y1, w, h = coords[:4]
bboxes.append([int(x1), int(y1), int(w), int(h)])
return bboxes
# 读取图片列表
def get_image_paths(seq_path):
img_dir = os.path.join(seq_path, "img")
img_names = sorted(os.listdir(img_dir))
return [os.path.join(img_dir, n) for n in img_names]
# 画框 + 生成视频
def visualize():
seq_path = os.path.join(DATASET_ROOT, SEQUENCE_NAME)
txt_path = os.path.join(RESULT_ROOT, f"{SEQUENCE_NAME}.txt")
if not os.path.exists(txt_path):
print(f"错误：找不到结果文件 {txt_path}")
return
bboxes = load_results(txt_path)
img_paths = get_image_paths(seq_path)
# 视频设置
first_img = cv2.imread(img_paths[0])
h, w = first_img.shape[:2]
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter(SAVE_VIDEO_PATH, fourcc, 30, (w, h))
print(f"正在生成视频：{SAVE_VIDEO_PATH}")
for i, img_path in enumerate(img_paths):
img = cv2.imread(img_path)
if i >= len(bboxes):
break
x1, y1, w, h = bboxes[i]
x2 = x1 + w
y2 = y1 + h
# 画跟踪框
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 3)
cv2.putText(img, f"Frame: {i+1}", (20, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 255, 0), 3)
out.write(img)
out.release()
print(f"✅ 视频已保存到：{SAVE_VIDEO_PATH}")
if __name__ == "__main__":
visualize()
第5步 生成auc
要进行修改后同第4步放在根目录下
import os
import numpy as np
# ===================== 【完全按你的真实路径填写】 =====================
# 你的预测结果路径
RESULT_DIR = r"D:\Program\test11\OSTrack-main\output\test\tracking_results\ostrack\vitb_384_mae_ce_32x4_ep300\lasot"
# 你的数据集真实标签路径（完全匹配你给的路径）
DATASET_ROOT = r"D:\Program\test11\OSTrack-main\data\lasot"
# ====================================================================
# 读取真实框 GT
def load_gt(seq_name):
class_name = seq_name.split('-')[0]
gt_path = os.path.join(DATASET_ROOT, class_name, seq_name, "groundtruth.txt")
gt = []
with open(gt_path, 'r') as f:
for line in f:
line = line.strip().replace(',', ' ')
parts = list(map(float, line.split()))
if len(parts) >= 4:
gt.append(parts[:4])
return np.array(gt)
# 读取预测框
def load_pred(seq_name):
pred_path = os.path.join(RESULT_DIR, f"{seq_name}.txt")
pred = []
with open(pred_path, 'r') as f:
for line in f:
line = line.strip().replace(',', ' ')
parts = list(map(float, line.split()))
if len(parts) >= 4:
pred.append(parts[:4])
return np.array(pred)
# IoU 计算
def iou(boxA, boxB):
x1, y1, w1, h1 = boxA
x2, y2, w2, h2 = boxB
xx1 = max(x1, x2)
yy1 = max(y1, y2)
xx2 = min(x1 + w1, x2 + w2)
yy2 = min(y1 + h1, y2 + h2)
w = max(0, xx2 - xx1)
h = max(0, yy2 - yy1)
inter = w * h
union = w1 * h1 + w2 * h2 - inter
return inter / (union + 1e-8)
# 计算标准 AUC
def compute_auc(ious):
thresholds = np.arange(0.0, 1.01, 0.01)
success = [np.mean(ious >= t) for t in thresholds]
return np.mean(success)
# 主函数
def main():
seq_list = [f[:-4] for f in os.listdir(RESULT_DIR) if f.endswith('.txt')]
print(f"✅ 找到序列总数：{len(seq_list)}")
auc_list = []
for seq in seq_list:
try:
gt = load_gt(seq)
pred = load_pred(seq)
except:
print(f"⚠️ {seq} 跳过")
continue
min_len = min(len(gt), len(pred))
ious = [iou(gt[i], pred[i]) for i in range(min_len)]
auc = compute_auc(ious)
auc_list.append(auc)
print(f"{seq:<28} | AUC = {auc:.4f}")
print("\n" + "=" * 70)
print("📊 所有序列 【标准平均 Success AUC】")
print("=" * 70)
print(f"✅ 平均 AUC = {np.mean(auc_list):.4f}")
print("=" * 70)
if __name__ == "__main__":
main()
生成可视化
import os
import numpy as np
import matplotlib.pyplot as plt
# ===================== 你的路径 =====================
RESULT_DIR = r"D:\Program\test11\OSTrack-main\output\test\tracking_results\ostrack\vitb_384_mae_ce_32x4_ep300\lasot"
DATASET_ROOT = r"D:\Program\test11\OSTrack-main\data\lasot"
# ====================================================
def load_gt(seq):
prefix = seq.split('-')[0]
path = os.path.join(DATASET_ROOT, prefix, seq, "groundtruth.txt")
gt = []
for line in open(path, errors='ignore'):
line = line.strip().replace(',', ' ')
vals = list(map(float, line.split()))
if len(vals) >= 4:
gt.append(vals[:4])
return gt
def load_pred(seq):
path = os.path.join(RESULT_DIR, seq + ".txt")
pred = []
for line in open(path, errors='ignore'):
line = line.strip().replace(',', ' ')
vals = list(map(float, line.split()))
if len(vals) >= 4:
pred.append(vals[:4])
return pred
def iou(a, b):
x1, y1, w1, h1 = a
x2, y2, w2, h2 = b
xx1 = max(x1, x2)
yy1 = max(y1, y2)
xx2 = min(x1 + w1, x2 + w2)
yy2 = min(y1 + h1, y2 + h2)
w = max(0, xx2 - xx1)
h = max(0, yy2 - yy1)
inter = w * h
union = w1 * h1 + w2 * h2 - inter
return inter / (union + 1e-8)
# ===================== 绘图主函数 =====================
def plot_curve():
seqs = [f[:-4] for f in os.listdir(RESULT_DIR) if f.endswith('.txt')]
all_ious = []
for seq in seqs:
try:
gt = load_gt(seq)
pred = load_pred(seq)
n = min(len(gt), len(pred))
for i in range(n):
all_ious.append(iou(gt[i], pred[i]))
except:
continue
thresholds = np.arange(0.0, 1.01, 0.01)
success = [np.mean(np.array(all_ious) >= t) for t in thresholds]
auc = np.mean(success)
plt.figure(figsize=(7, 5))
plt.plot(thresholds, success, 'b-', linewidth=3, label=f'AUC = {auc:.3f}')
plt.fill_between(thresholds, success, alpha=0.3, color='blue')
plt.xlabel('IoU Threshold')
plt.ylabel('Success Rate')
plt.title('LaSOT Success Curve (AUC)')
plt.grid(True)
plt.legend()
plt.savefig('auc_curve.png', dpi=300)
plt.close()
print(f"✅ 图片已生成：auc_curve.png")
print(f"✅ 最终 AUC = {auc:.3f}")
if __name__ == '__main__':
plot_curve()
第6步 otb100数据集
修改路径
打开OSTrack-main\lib\test\evaluation\local.py
settings.otb_path = 'D:/Program/test11/OSTrack-main/data/OTB100'改成实际路径
打开lib/test/evaluation/otbdataset.py
40行左右ground_truth_rect = load_text(str(anno_path), delimiter=(',', None), dtype=np.float64, backend='numpy')
改成ground_truth_rect = load_text(str(anno_path), delimiter=',', dtype=np.float64, backend='numpy')
打开lib/test/utils/load_text.py注释
ground_truth_rect = np.loadtxt(path, delimiter=d, dtype=dtype)换成
with open(path, 'r') as f:
ground_truth_rect=np.loadtxt(io.StringIO(f.read().replace(',', ' ')))
打开
lib/test/utils/load_text.py直接全部改成（豆包出品）
import numpy as np
import pandas as pd
import io
def load_text_numpy(path, delimiter, dtype):
try:
with open(path, 'r', encoding='utf-8', errors='ignore') as f:
lines = []
for line in f:
line = line.strip()
if not line:
continue
line = line.replace(',', ' ')
parts = list(map(float, line.split()))
if len(parts) >= 4:
lines.append(parts[:4])
return np.array(lines, dtype=dtype)
except:
pass
raise Exception('Could not read file {}'.format(path))
def load_text_pandas(path, delimiter, dtype):
try:
with open(path, 'r', encoding='utf-8', errors='ignore') as f:
lines = []
for line in f:
line = line.strip()
if not line:
continue
line = line.replace(',', ' ')
parts = list(map(float, line.split()))
if len(parts) >= 4:
lines.append(parts[:4])
return np.array(lines, dtype=dtype)
except:
pass
raise Exception('Could not read file {}'.format(path))
def load_text(path, delimiter=' ', dtype=np.float32, backend='numpy'):
if backend == 'numpy':
return load_text_numpy(path, delimiter, dtype)
elif backend == 'pandas':
return load_text_pandas(path, delimiter, dtype)
def load_str(path):
with open(path, "r") as f:
text_str = f.readline().strip().lower()运行python tracking/test.py ostrack vitb_384_mae_ce_32x4_ep300 --dataset_name otb --threads 2 --num_gpus 1
更多推荐
12
0- 0
已为社区贡献8条内容
所有评论(0)