def process(task_id):
# get meta info
meta = TaskMeta.from_redis(task_id)
logging.info('[process] msg: %s', task_id)
if not meta:
logging.error('[process] get from_redis failed, task_id: %s', task_id)
return ERR_READ_REDIS_FAILED
# inference
result_img = inference(meta.input_url)
# upload result
obj_name = upload_img(meta.task_id, result_img)
logging.info('[process] upload_img: %s, obj_name: %s', task_id, obj_name)
if not obj_name:
logging.error('[process] upload_img failed, task_id: %s', task_id)
return ERR_WRITE_CLOUD_STORAGE_FAILED
output_url = URL % obj_name
# update meta info
meta.output_url = output_url
meta.proc_status = TaskMeta.TASK_STATUS_SUC
logging.info('[process] meta to update: %s', meta.to_dict())
if not meta.update():
logging.error('[process] meta.update failed, task_id: %s', task_id)
return ERR_WRITE_REDIS_FAILED
logging.info('[process] done task: %s', task_id)
return SUC
# lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=4, gamma=0.3)
scheduler_cosine = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, config.num_epochs*len(data_loader))
lr_scheduler = GradualWarmupScheduler(optimizer, multiplier=100,
total_epoch=min(1000, len(data_loader)-1),
after_scheduler=scheduler_cosine)
# loss function
criterion = DiceLoss()
# criterion = Weight_Soft_Dice_Loss(weight=[0.1, 0.9])
# criterion = BCELoss()
# criterion = MixedLoss(10.0, 2.0)
# criterion = Weight_BCELoss(weight_pos=0.25, weight_neg=0.75)
# criterion = Lovasz_Loss(margin=[1, 5]
print('start training...')
train_start = time.time()
for epoch in range(config.num_epochs):
epoch_start = time.time()
model_ft, optimizer = train_one_epoch(model_ft, data_loader, criterion,
optimizer, lr_scheduler=lr_scheduler, device=device,
epoch=epoch, vis=vis)
do_valid(model_ft, dataloader_val, criterion, epoch, device, vis=vis)
print('Epoch time: {:.3f}min\n'.format((time.time()-epoch_start)/60/60))
print('total train time: {}hours {}min'.format(int((time.time()-train_start)/60//60), int((time.time()-train_start)/60%60)))
inference_all(model_ft, device=device)
inference(model_ft, device=device)
torch.save(model_ft, f'{config.model}.pth')
torch.save({'optimizer': optimizer.state_dict(),
'epoch': epoch,},
'optimizer.pth')
from torch.utils.data import DataLoader, Dataset
from dataset import Image_Dataset
from sklearn.decomposition import KernelPCA
from sklearn.manifold import TSNE
from sklearn.cluster import MiniBatchKMeans, KMeans
from preprocess import preprocess
import torchvision.transforms as transforms
from test import inference, predict, save_prediction, invert
#load model
model = Improved_AE().cuda()
model.load_state_dict(torch.load(sys.argv[2]))
model.eval()
# 準備 data
trainX = np.load(sys.argv[1])
# 預測答案
latents = inference(X=trainX, model=model)
pred, X_embedded = predict(latents)
# 將預測結果存檔，上傳 kaggle
if pred[6] == 1:
save_prediction(pred, sys.argv[3])
# 由於是 unsupervised 的二分類問題，我們只在乎有沒有成功將圖片分成兩群
# 如果上面的檔案上傳 kaggle 後正確率不足 0.5，只要將 label 反過來就行了
else:
save_prediction(invert(pred), sys.argv[3])
import numpy as np
import torch
from test import inference, predict
from utils import cal_acc, plot_scatter
from autoencoder import Base_AE, Improved_AE
from preprocess import preprocess
valX = np.load('data/valX.npy')
valY = np.load('data/valY.npy')
#model = Base_AE().cuda()
model = Improved_AE().cuda()
model.load_state_dict(torch.load('./improve-2.pth'))
model.eval()
latents = inference(valX, model)
pred_from_latent, emb_from_latent = predict(latents)
acc_latent = cal_acc(valY, pred_from_latent)
print('The clustering accuracy is:', acc_latent)
print('The clustering result:')
plot_scatter(emb_from_latent, valY, savefig='p1_improved.png')
'''
import matplotlib.pyplot as plt
import numpy as np
# 畫出原圖
trainX = np.load('data/trainX_new.npy')
trainX_preprocessed = preprocess(trainX)
model = Improved_AE().cuda()
model.load_state_dict(torch.load('./improve-2.pth'))
import numpy as np
if __name__=='__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-src', required=True)
parser.add_argument('-model', required=True, help='path to pre-trained model')
parser.add_argument('-output', required=True, help='path to store the output')
parser.add_argument('-batch', type=int, default=32, required=True, help='size of a batch')
opt = parser.parse_args()
dataset=make_src_dataset(opt.src, opt.batch)
iterator=dataset.make_initializable_iterator()
src_input,src_size=iterator.get_next()
#定义计算图
sequence_op = inference(src_input,src_size)
saver = tf.train.Saver()
result = []
with tf.Session() as sess:
saver.restore(sess, opt.model)
sess.run(iterator.initializer)
print('[INFO] Start inference process...')
while True:
step = 0
try:
sequence = sess.run(sequence_op).tolist()
step += 1
if (step * opt.batch) % 100 ==0:
print('{} test instances finished.'.format(step*opt.batch))
for seq in sequence:
def upload():
global shot_type
global destination
global batsman
if request.method == 'POST':
f = request.files['video-input']
# This will contain the name of the batsman selected form the dropdown
action_type = request.form['action-type']
print("Action type is:", action_type)
batsman = request.form['batsman']
# This will contain the path of the file selected
destination = "\\".join([target, f.filename])
f.save(destination)
print(destination)
if action_type == 'analyze':
# Analyze logic goes here
preprocess(destination)
bowl_ids=inference()
#return str(destination)+" "+str(batsman)
#changing variables
bm=batsman
bi=bowl_ids
print("initial bowling ids predicted by model",bi)
x=bi[0]#1st bowl id
y=bi[1]#2nd bowl id
#after manipulation
a=map[x]#1st bowl id
b=map[y]#2nd bowl # IDEA:
bi[0]=a
bi[1]=b
print("final bowling ids after manipulation according to dataset",bi)
#most played shot
mps=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
mps=mps.mode()['Shot_Played'][0]
#ideal shot
ids=df.loc[ (df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1])]
ids=ids.groupby('Shot_Played').mean().reset_index()
ids=ids.iloc[ids["Boundary_Success_Rate"].idxmax()]
ids=ids["Shot_Played"]
#avg power
ap=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
ap=ap["Power_Applied(N)"].mean()
#avg reaction time
art=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
art=art["Reaction Time(sec)"].mean()
#best refrence
br=df.loc[ (df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1])]
br=br.groupby('Batsman').mean().reset_index()
br=br.iloc[br["Boundary_Success_Rate"].idxmax()]
br=br["Batsman"]
#return str(destination)+" "+str(batsman)
stats = {
'Most Played Shot': mps,
'Ideal Shot': ids,
'Average Power': ap,
'Average Reaction Time': art,
'Best Refrence': br,
}
'''
stats = {
'Most Played Shot': 'Cover drive',
'Ideal Shot': 'Sixer',
'Average Power': '12 W',
'Average Reaction Time': '100 ms'
}
'''
dir_path = 'C:/Users/Administrator/Desktop/Ankita/uploads/'
try:
shutil.rmtree(dir_path)
except OSError as e:
print("Error: %s : %s" % (dir_path, e.strerror))
return render_template('stats.html', batsman = batsman, stats = stats)
elif action_type == 'coachme':
print(batsman)
print(shot_type)
# Coach me logic goes here
Cover_drive=[[-0.5, 1.2, -9.1,0,-1,0],[-0.55, 1.2, -8.5,0,-1,0],[-0.58, 1.2, -8.3,0,-1,0],[-0.6, 1.2, -8.0,0,-1,0],[-0.65, 1.2, -7.8,-15,-1,0],[-0.68, 1.2, -7.7,-19,-1,0],[-0.7, 1.2, -7.5,-20,-1,0],[-0.72, 1.2, -7.3,-25,-1,0]]
Straight_drive=[[-0.5, 1.2, -9.1,0,0,0],[-0.5, 1.2, -8.5,0,0,0],[-0.5, 1.2, -8.3,0,0,0],[-0.5, 1.2, -8.0,0,0,0],[-0.5, 1.2, -7.8,0,0,0],[-0.5, 1.2, -7.7,-20,0,0],[-0.5, 1.2, -7.5,-29,0,0],[-0.5, 1.2, -7.3,-30,0,0]]
if batsman=="Hari":
shot_type=Straight_drive
if batsman=="Vivek":
shot_type=Cover_drive
return render_template('coach.html',shot_type=shot_type)
print("Batsman:", batsman)
print("Filename", destination)
return render_template('index.html', batsmen=batsmen, len=len(batsmen), batsman=batsman, destination=destination)
def analyse():
global selected_player
global selected_delivery
global destination
global batsman
if request.method == 'POST':
print(request.json)
selected_player = request.json['player']
selected_delivery = request.json['delivery']
print("Selected player in POST is: ", selected_player)
print("Player details: ", selected_player['id'], selected_player['name'], selected_player['imageUrl'], selected_player['isCaptain'])
print("Delivery details: ", selected_delivery['id'], selected_delivery['name'], selected_delivery['videoUrl'])
# Player and delivery details are available in the variable above, as described
resp = jsonify(success=True)
return resp
if request.method == 'GET':
batsman=selected_player['name']
original="C:/Users/Administrator/Desktop/Cricket_new_UI/" + selected_delivery['videoUrl']
shutil.copyfile(original, 'C:/Users/Administrator/Desktop/Ankita/uploads/test.webm')
print("Selected player in GET is: ", selected_player)
print("Selected delivery in GET is: ", selected_delivery['videoUrl'])
destination = "C:\\Users\\Administrator\\Desktop\\Ankita\\uploads\\test.webm"
# Analyze logic goes here
preprocess(destination)
bowl_ids=inference()
#return str(destination)+" "+str(batsman)
#changing variables
bm=batsman
bi=bowl_ids
print("initial bowling ids predicted by model",bi)
x=bi[0]#1st bowl id
y=bi[1]#2nd bowl id
#after manipulation
a=map[x]#1st bowl id
b=map[y]#2nd bowl # IDEA:
bi[0]=a
bi[1]=b
print("final bowling ids after manipulation according to dataset",bi)
#most played shot
mps=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
mps=mps.mode()['Shot_Played'][0]
#ideal shot
ids=df.loc[ (df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1])]
ids=ids.groupby('Shot_Played').mean().reset_index()
ids=ids.iloc[ids["Boundary_Success_Rate"].idxmax()]
ids=ids["Shot_Played"]
#avg power
ap=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
ap=ap["Power_Applied(N)"].mean()
#avg reaction time
art=df.loc[(df['Batsman'] == bm) & ((df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1]))]
art=art["Reaction Time(sec)"].mean()
#best refrence
br=df.loc[ (df['Bowl_Id'] == bi[0]) | (df['Bowl_Id'] == bi[1])]
br=br.groupby('Batsman').mean().reset_index()
br=br.iloc[br["Boundary_Success_Rate"].idxmax()]
br=br["Batsman"]
#return str(destination)+" "+str(batsman)
stats = {
'Most Played Shot': mps,
'Ideal Shot': ids,
'Average Power': ap,
'Average Reaction Time': art,
'Best Refrence': br,
}
'''
stats = {
'Most Played Shot': 'Cover drive',
'Ideal Shot': 'Sixer',
'Average Power': '12 W',
'Average Reaction Time': '100 ms'
}
'''
dir_path = 'C:/Users/Administrator/Desktop/Ankita/uploads/'
try:
shutil.rmtree(dir_path)
except OSError as e:
print("Error: %s : %s" % (dir_path, e.strerror))
try:
os.makedirs(dir_path)
except OSError as e:
pass
return render_template('stats.html', batsman = batsman, stats = stats)
#dataset = Image_Dataset(trainX, test_transform)
dataloader = DataLoader(dataset, batch_size=64, shuffle=False)
model = Improved_AE().cuda()
points = []
with torch.no_grad():
for i, checkpoint in enumerate(checkpoints_list):
print('[{}/{}] {}'.format(i + 1, len(checkpoints_list), checkpoint))
model.load_state_dict(torch.load(checkpoint))
model.eval()
err = 0
n = 0
for x in dataloader:
x = x.cuda()
_, rec = model(x)
err += torch.nn.MSELoss(reduction='sum')(x, rec).item()
n += x.flatten().size(0)
print('Reconstruction error (MSE):', err / n)
latents = inference(X=valX, model=model)
pred, X_embedded = predict(latents)
acc = cal_acc(valY, pred)
print('Accuracy:', acc)
points.append((err / n, acc))
'''
ps = list(zip(*points))
plt.figure(figsize=(6,6))
plt.subplot(211, title='Reconstruction error (MSE)').plot(epoch, ps[0])
plt.subplot(212, title='Accuracy (val)', xlabel='Epoch').plot(epoch, ps[1])
plt.show()
'''
def train():
x = tf.placeholder(tf.float32,
shape=[
BATCH_SIZE, test.IMAGE_SIZE, test.IMAGE_SIZE,
test.NUM_CHANNELS
],
name='x_input')
y_ = tf.placeholder(tf.float32,
shape=[BATCH_SIZE, test.OUTPUT_NODE],
name='y_input')
regularizer = tf.contrib.layers.l2_regularizer(REGULARIZATION_RATE)
y = test.inference(x, train, regularizer=regularizer)
global_step = tf.Variable(0, trainable=False)
variable_averages = tf.train.ExponentialMovingAverage(
MOVING_AVERAGE_DECAY, global_step)
variable_averages_op = variable_averages.apply(tf.trainable_variables())
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=y, labels=tf.argmax(y_, 1))
cross_entropy_mean = tf.reduce_mean(cross_entropy)
loss = cross_entropy_mean + tf.add_n(tf.get_collection("losses"))
learning_rate = tf.train.exponential_decay(LEARNING_RATE_BASE, global_step,
TRAINING_STEPS / BATCH_SIZE,
LEARNING_RATE_DECAY)
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(
loss, global_step)
with tf.control_dependencies([train_step, variable_averages_op]):
train_op = tf.no_op(name='Train')
#初始化tensorflow持久化类
# saver=tf.train.Saver()
with tf.Session() as sess:
tf.global_variables_initializer().run()
####--------------------------设置BATCH_SIZE------------------------####
for i in range(TRAINING_STEPS):
ID = np.random.randint(0, 199, BATCH_SIZE)
xs = test.train_data[ID]
ys = test.train_label[ID]
_, loss_value, step = sess.run([train_op, loss, global_step],
feed_dict={
x: xs,
y_: ys
})
if i % 50 == 0:
print(
"After %d steps training steps,loss on training batch is %g"
% (step, loss_value))
# saver.save(sess,os.path.join(MODEL_SAVE_PATH,MODEL_NAME),global_step=global_step)
ypre = sess.run(y, feed_dict={x: test_data})
y1 = np.argmax(ypre)
y2 = np.argmax(test_label)
accuracy = np.sum(y1 - y2)
accuracy = accuracy / numtest
print('Accuracy_learn ============>> ', accuracy)
def test():
input_url = 'https://storage.googleapis.com/ylq_server/208e8322d7ecd9f48c969be0c95b333e'
result_img = inference(input_url)
# upload result
result_img.save('/tmp/res', 'JPEG')