FK Crvena Zvezda contra FK Radnik Surdulica

FK Crvena Zvezda vs FK Radnik Surdulica: Expert Analysis and Predictions

The upcoming football match between FK Crvena Zvezda and FK Radnik Surdulica promises to be an exciting encounter, with both teams eager to claim victory on November 2, 2025, at 16:00. FK Crvena Zvezda, the home team, is favored to win with odds of 94.50, reflecting their strong performance and home advantage. The betting odds suggest a high-scoring game, with the average total goals predicted at 4.36, indicating an action-packed match. Both teams have shown their offensive capabilities, with an average of 2.75 goals scored per game by the home team and 1.71 conceded goals by the away team. The possibility of a red card is also notable, with an average of 1.49 cards per match.

Predictions for Key Betting Lists

liliang2016/2017_10_24/BBDNN/keras/dataGenerator.py
from keras.preprocessing.image import ImageDataGenerator
import numpy as np

def dataGen(batch_size=16):

gen = ImageDataGenerator(
rotation_range=45,
width_shift_range=0.15,
height_shift_range=0.15,
zoom_range=0.15,
horizontal_flip=True,
vertical_flip=False,
fill_mode=’reflect’
)

return gen.flow_from_directory(
‘data/train’,
target_size=(256,256),
batch_size=batch_size,
class_mode=’categorical’,
shuffle=True)

def dataGenTest(batch_size=16):

return ImageDataGenerator().flow_from_directory(
‘data/test’,
target_size=(256,256),
batch_size=batch_size,
class_mode=None,
shuffle=False)

def dataGenPredict(batch_size=16):

return ImageDataGenerator().flow_from_directory(
‘data/predict’,
target_size=(256,256),
batch_size=batch_size,
class_mode=None,
shuffle=False)

if __name__ == ‘__main__’:

train_generator = dataGen()
test_generator = dataGenTest()
predict_generator = dataGenPredict()

print(train_generator.next())
print(test_generator.next())
print(predict_generator.next())#coding=utf-8
from __future__ import division
import tensorflow as tf
import numpy as np
import cv2
import os
import math

def randomCrop(imgs,w,h,crop_h,crop_w):

# x,y=random.randint(0,w-crop_w),random.randint(0,h-crop_h)
# x,y=random.randint(0,w-crop_w),random.randint(0,h-crop_h)
# return imgs[:,x:x+crop_w,y:y+crop_h,:]

# w_cropped=int(w*0.8)
# h_cropped=int(h*0.8)
# x=random.randint(0,w-w_cropped)
# y=random.randint(0,h-h_cropped)
# return imgs[:,x:x+w_cropped,y:y+h_cropped,:]

# w_cropped=int(w*random.uniform(0.6,1))
# h_cropped=int(h*random.uniform(0.6,1))
# x=random.randint(0,w-w_cropped)
# y=random.randint(0,h-h_cropped)
# return imgs[:,x:x+w_cropped,y:y+h_cropped,:]

# w_cropped=int(w*random.uniform(0.,1))
# h_cropped=int(h*random.uniform(0.,1))
# x=random.randint(0,w-w_cropped)
# y=random.randint(0,h-h_cropped)
# return imgs[:,x:x+w_cropped,y:y+h_cropped,:]

w_cropped=int(w*random.uniform(0.,1))
h_cropped=int(h*random.uniform(0.,1))

if w_cropped>=crop_w and h_cropped>=crop_h:

x=random.randint(0,w-w_cropped)
y=random.randint(0,h-h_cropped)

return imgs[:,x:x+crop_w,y:y+crop_h,:]

else:

while True:

x=random.randint(0,w-crop_w)
y=random.randint(0,h-crop_h)

if imgs[2,x:x+crop_w,y:y+crop_h].mean()>100:

break

return imgs[:,x:x+crop_w,y:y+crop_h,:]

def randomFlip(imgs):

# if random.random()<0.5:

# for i in range(imgs.shape[0]):

# imgs[i,:,:,:]=cv2.flip(imgs[i,:,:,:],1)

# return imgs

# else:

# return imgs

# if random.random()<0.5:

# for i in range(imgs.shape[0]):

# imgs[i,:,:,:]=cv2.flip(imgs[i,:,:,:],-1)

# return imgs

# else:

# return imgs

def preprocess(imgs,batch_size):

def procTrain(imgs,batch_size):

cropped_imgs=np.zeros([batch_size]+list(imgs.shape[1:]),dtype=np.uint8)

for i in range(batch_size):

cropped_imgs[i]=randomCrop(imgs[i],imgs.shape[2],imgs.shape[3],imgs.shape[2],imgs.shape[3])

imgs=cropped_imgs

cropped_imgs=np.zeros([batch_size]+list(imgs.shape[1:]),dtype=np.uint8)

for i in range(batch_size):

cropped_imgs[i]=randomCrop(imgs[i],imgs.shape[2],imgs.shape[3],128,128)

imgs=cropped_imgs

def procValid(imgs,batch_size):

cropped_imgs=np.zeros([batch_size]+list(imgs.shape[1:]),dtype=np.uint8)

for i in range(batch_size):

cropped_imgs[i]=cv2.resize(imgs[i],dsize=(128,128))

imgs=cropped_imgs

def procTest(imgs,batch_size):

cropped_imgs=np.zeros([batch_size]+list(imgs.shape[1:]),dtype=np.uint8)

for i in range(batch_size):

cropped_imgs[i]=cv2.resize(imgs[i],dsize=(128,128))

imgs=cropped_imgs

if __name__ == '__main__':

liliang2016/2017_10_24<|file_sep因为使用了不同的图片数据集，所以这两个版本的训练结果差别比较大。

运行方法：

python main.py -gpu [gpu_id] -num_epoch [num_epoch] -batch [batch size] -lr [learning rate] -model [model name] -train [train or test]

具体选项：

–gpu : 使用哪个GPU，例如： –gpu=4，如果要使用CPU训练则设为–gpu=-1.

–num_epoch : 训练轮数，一般设置为500，这里我已经将代码中的epoch设置为500了，所以可以不用指定。

–batch : 批大小，一般设置为32。

–lr : 学习率，一般设置为10e-4.

–model : 模型名称，可选的有： BBDNN、BBDNN_noBN、BBDNN_dropout、BBDNN_dropBN、BBDNN_dropBN_bn。具体的模型可以在main.py中看到。

–train : 是否是训练阶段，即是否要保存模型。如果只是测试阶段，则指定该参数为test，否则会保存训练过程中生成的模型。

使用说明：

首先将要用到的模型和数据放在同一个文件夹下，并且需要在main.py中指定该文件夹路径。

然后将train.py和test.py放在同一个文件夹下，并且在main.py中指定该文件夹路径。

最后使用python main.py + 具体选项进行训练或测试。

运行时注意事项：

如果要使用GPU训练，则需要在运行前执行下面的命令：

CUDA_VISIBLE_DEVICES=[gpu id] python main.py + 具体选项。

如果要使用CPU训练，则直接执行下面的命令：

python main.py + 具体选项。<|file_sep flycatcher/dataset/UCF101/UCF-101/images/
flycatcher/dataset/UCF101/UCF-101/train_list.txt
flycatcher/dataset/UCF101/UCF-101/test_list.txtliliang2016/2017_10_24= v1.x

### **Training**

For training our model on UCF101 dataset (with spatial-temporal jittering), run:

`python train.py –train_dir=`

For training our model on HMDB51 dataset (with spatial-temporal jittering), run:

`python train.py –train_dir=`

For training our model on HMDB51 dataset (without spatial-temporal jittering), run:

`python train.py –train_dir=`

where « should point to folder containing subfolders named ‘images’ and ‘annotations’.

### **Testing**

For testing our model on UCF101 dataset (with spatial-temporal jittering), run:

`python test.py –test_dir=`

For testing our model on HMDB51 dataset (with spatial-temporal jittering), run:

`python test.py –test_dir=`

For testing our model on HMDB51 dataset (without spatial-temporal jittering), run:

`python test.py –test_dir=`

where « should point to folder containing subfolders named ‘images’ and ‘annotations’.

### **Model**

Our pre-trained models can be downloaded from https://drive.google.com/drive/folders/13kGjg8uyWkeN7YfXvZPXyAwyd9QnIbTm?usp=sharing.

liliang2016/2017_10_24<|file_sepusually you should change these path before using this code:

datasetPath = '/home/flycatcher/dataset/Something-Something-V1/'

savePath = '/home/flycatcher/exp/mv'

modelPath = '/home/flycatcher/exp/mv/model/'liliang2016/2017_10_24<|file_sep whoosh==2.7.* # For text search index.
scikit-image==0.* # For image processing.
matplotlib==2.* # For plotting.
scikit-learn==0.* # For computing accuracy metrics.
pyyaml==4.* # For configuration files.
pillow==4.* # For image processing.
scipy==1.* # For signal processing.
tensorflow-gpu==1.*
keras==2.*
tqdm==4.*
opencv-python==4.*
scikit-video==1.*
imutils==0.*
numpy==1.*
PIL==6.*

liliang2016/2017_10_24<|file_sepsys.path.append('D:/dataset/')
sys.path.append('D:/dataset/code/')
sys.path.append('D:/dataset/code/codes_for_papers/BBDNN/')
sys.path.append('D:/dataset/code/codes_for_papers/BBDNN/tensorflow/')
sys.path.append('D:/dataset/code/codes_for_p