def Only_Merge_Spatial_Net_Test():
spa_model = resnet152_SpatialNet().cuda().eval()
if spa_model_save_file is not None:
spa_model.load_state_dict(torch.load(spa_model_save_file))
print('load spa_model success!')
dsl = test_UCF101_ChooseRandomFromSameVideo(dsl=UCF101_Spatial)
# dsl = test_UCF0101_Spatial()
# def gen():
# return GenVariables_Spatial(dsl,batchsize=16)
loops = 500
correct_1 = 0
correct_5 = 0
correct_10 = 0
for l in range(loops):
imgpaths, Lables = random.choice(dsl)
imgs = []
for path in imgpaths:
imgs.append(cv2.imread(path))
imgs = np.array(imgs)
imgs = ImgAugPipes(imgs, NoAug=True, isNormal=True)
imgs = Variable(torch.from_numpy(imgs)).cuda().float()
lable = Variable(torch.from_numpy(np.array(Lables))).cuda().long()
pred = spa_model(imgs[:3])
lable = Variable(torch.from_numpy(np.array([Lables[0]]))).cuda().long()
pred = pred.sum(0) / 3
# acc = accuracy(pred.cpu(),lable.cpu(),(1,5,10))
acc = accuracy(pred.cpu(), lable.cpu(), (1, 5, 10))
correct_1 += acc[0]
correct_5 += acc[1]
correct_10 += acc[2]
print(l, ':', acc)
print('acc@1:', correct_1 / loops)
print('acc@5:', correct_5 / loops)
print('acc@10:', correct_10 / loops)
'''
def Only_Merge_Temporal_Net():
dsl = test_UCF101_ChooseRandomFromSameVideo(dsl=UCF101_TwoStream)
tem_model = resnet152_TemporalNet().cuda()
if tem_model_save_file is not None:
tem_model.load_state_dict(torch.load(tem_model_save_file))
print('load tem_model success!')
tem_model.eval()
loops = 100
correct_1 = 0
correct_5 = 0
correct_10 = 0
for l in range(loops):
imgpathss, labels = random.choice(dsl)
imgpathss = [imgpaths[1:] for imgpaths in imgpathss]
imgs = GenTensors(imgpathss, isTemporal=True, NoAug=True)
imgs = Variable(imgs).cuda()
labels = Variable(torch.from_numpy(np.array(labels))).cuda().long()
pred = tem_model(imgs)
# acc = accuracy(pred,labels,topk=(1,5,10))
pred = pred.sum(0) / 8
lable = labels[0].cuda().long()
acc = accuracy(pred, lable, topk=(1, 5, 10))
print(l, ':', acc)
correct_1 += acc[0]
correct_5 += acc[1]
correct_10 += acc[2]
print('acc@1:', correct_1 / loops)
print('acc@5:', correct_5 / loops)
print('acc@10:', correct_10 / loops)
def TestOnlyTemporalNet():
tem_model = resnet152_TemporalNet().cuda()
if tem_model_save_file is not None:
tem_model.load_state_dict(torch.load(tem_model_save_file))
print('load tem_model success!')
tem_model.eval()
dsl = test_UCF0101_Temporal()
def gen():
return GenVariables_Temporal(dsl, batchsize=16)
loops = 10
correct_1 = 0
correct_5 = 0
correct_10 = 0
for l in range(loops):
images, labels = gen()
pred = tem_model(images)
# pred = F.softmax(pred)
labels = labels.cpu()
pred = pred.cpu()
acc = accuracy(pred, labels, topk=(1, 5, 10))
correct_1 += acc[0]
correct_5 += acc[1]
correct_10 += acc[2]
print(l, ':', acc)
print('acc@1: ', correct_1 / loops)
print('acc@5: ', correct_5 / loops)
print('acc@10: ', correct_10 / loops)
'''
def C3D_Net_Run():
epochs = 81
loops = 2001
learningrate = 0.0001
attenuation = 0.1
model = C3D(drop=0.9).cuda()
if Config.LOAD_SAVED_MODE_PATH is not None:
import types
model.try_to_load_state_dict = types.MethodType(
try_to_load_state_dict, model)
model.try_to_load_state_dict(torch.load(Config.LOAD_SAVED_MODE_PATH))
print('LOAD {} done!'.format(Config.LOAD_SAVED_MODE_PATH))
lossfunc = nn.CrossEntropyLoss()
optim = torch.optim.Adam(model.parameters(), lr=learningrate)
pq_train = PictureQueue(dsl=train_UCF101_C3D(),
Gen=GenVariables_C3D,
batchsize=batchsize,
worker=5)
pq_test = PictureQueue(dsl=test_UCF101_C3D(),
Gen=GenVariables_C3D,
batchsize=batchsize,
worker=2)
cnt = 0
for epoch in range(epochs):
for l in range(loops):
cnt += 1
imgs, labels = pq_train.Get()
model.zero_grad()
pred = model(imgs)
loss = lossfunc(pred, labels)
logger.scalar_summary('C3D/train_loss', loss.data[0], cnt)
loss.backward()
optim.step()
print('C3D epoch: {} cnt: {} loss: {}'.format(
epoch, cnt, loss.data[0]))
if cnt % 25 == 0:
imgs, labels = pq_test.Get()
pred = model.inference(imgs)
loss = lossfunc(pred, labels)
logger.scalar_summary('C3D/test_loss', loss.data[0], cnt)
# acc
acc = accuracy(pred, labels, topk=(1, 5, 10))
logger.scalar_summary('C3D/test_acc@1', acc[0], cnt)
logger.scalar_summary('C3D/test_acc@5', acc[1], cnt)
logger.scalar_summary('C3D/test_acc@10', acc[2], cnt)
imgs, labels = pq_train.Get()
pred = model.inference(imgs)
acc = accuracy(pred, labels, topk=(1, 5, 10))
logger.scalar_summary('C3D/train_acc@1', acc[0], cnt)
logger.scalar_summary('C3D/train_acc@5', acc[1], cnt)
logger.scalar_summary('C3D/train_acc@10', acc[2], cnt)
if cnt % 2000 == 0:
savefile = savepath + 'C3D_EX1_{:02d}.pt'.format(epoch % 20)
print('C3D save model to {}'.format(savefile))
torch.save(model.state_dict(), savefile)
if epoch in [20, 40, 60]:
learningrate = learningrate * attenuation
optim = torch.optim.Adam(model.parameters(), lr=learningrate)
def Merge_Test_2():
spa_model = resnet152_SpatialNet().cuda()
tem_model = resnet152_TemporalNet().cuda()
if spa_model_save_file is not None:
spa_model.load_state_dict(torch.load(spa_model_save_file))
print('load spa_model success!')
if tem_model_save_file is not None:
tem_model.load_state_dict(torch.load(tem_model_save_file))
print('load tem_model success!')
spa_model.eval()
tem_model.eval()
dsl = test_UCF101_ChooseRandomFromSameVideo(dsl=UCF101_TwoStream)
loops = len(dsl)
correct_1 = 0
correct_5 = 0
correct_10 = 0
f = open('/tmp/test_log.txt', 'w')
for l in range(loops):
# imgpathss,labels = random.choice(dsl)
imgpathss, labels = dsl[l]
imgpaths_s = [imgpaths[0] for imgpaths in imgpathss]
imgpaths_t = [imgpaths[1:] for imgpaths in imgpathss]
imgs_t = GenTensors(imgpaths_t, isTemporal=True, NoAug=True)
imgs_t = Variable(imgs_t).cuda()
labels = Variable(torch.from_numpy(np.array(labels))).cuda().long()
line = '{},{},'.format(l, labels[0].cpu().data.numpy()[0])
pred_t = tem_model(imgs_t)
# acc = accuracy(pred,labels,topk=(1,5,10))
pred_t = pred_t.sum(0) / 8
lable = labels[0].cuda().long()
acc = accuracy(pred_t, lable, topk=(1, 5, 10))
line += str(pred_t.cpu().data.numpy().tolist()[0])[1:-1] + ","
print('l:', l)
print('temp :', acc)
imgs_s = []
for path in imgpaths_s:
imgs_s.append(cv2.imread(path))
imgs_s = np.array(imgs_s)
imgs_s = ImgAugPipes(imgs_s, NoAug=True, isNormal=True)
imgs_s = Variable(torch.from_numpy(imgs_s)).cuda().float()
pred_s = spa_model(imgs_s)
pred_s = pred_s.sum(0) / 8
line += str(pred_s.cpu().data.numpy().tolist()[0])[1:-1]
acc = accuracy(pred_s, lable, topk=(1, 5, 10))
print('spa:', acc)
pred_a = (pred_s + pred_t) / 2
acc = accuracy(pred_a, lable, topk=(1, 5, 10))
print('all:', acc)
correct_1 += acc[0]
correct_5 += acc[1]
correct_10 += acc[2]
if l % 50 == 0:
print('acc@1:', correct_1 / (l + 1))
print('acc@5:', correct_5 / (l + 1))
print('acc@10:', correct_10 / (l + 1))
f.write(line + '\n')
f.flush()
print('acc@1:', correct_1 / loops)
print('acc@5:', correct_5 / loops)
print('acc@10:', correct_10 / loops)
def VGG_Spatial_Net_Run():
epochs = 80
loops = 2000
learningrate = 0.001
attenuation = 0.1
model = VGG_Spatial_Net(pretrained=False, dropout1=0.8, dropout2=0.7).cuda()
if Config.LOAD_SAVED_MODE_PATH is not None:
import types
model.try_to_load_state_dict = types.MethodType(try_to_load_state_dict, model)
model.try_to_load_state_dict(torch.load(Config.LOAD_SAVED_MODE_PATH))
print('LOAD {} done!'.format(Config.LOAD_SAVED_MODE_PATH))
lossfunc = nn.CrossEntropyLoss()
optim = torch.optim.SGD(model.parameters(), lr=learningrate, momentum=0.9)
cnt = 0
pq_train = PictureQueue(dsl=train_UCF0101_Spatial(), Gen=GenVariables_Spatial, batchsize=batchsize)
pq_test = PictureQueue(dsl=test_UCF0101_Spatial(), Gen=GenVariables_Spatial, batchsize=batchsize)
for epoch in range(epochs):
for l in range(loops):
cnt += 1
imgs, labels = pq_train.Get()
model.zero_grad()
pred = model(imgs)
loss = lossfunc(pred, labels)
logger.scalar_summary('Spatial/train_loss', loss.data[0], cnt)
loss.backward()
optim.step()
print('Spatial epoch: {} cnt: {} loss: {}'.format(epoch, cnt, loss.data[0]))
if cnt % 20 == 0:
imgs, labels = pq_test.Get()
pred = model.inference(imgs)
loss = lossfunc(pred, labels)
logger.scalar_summary('Spatial/test_loss', loss.data[0], cnt)
acc = accuracy(pred, labels, topk=(1, 5, 10))
logger.scalar_summary('Spatial/test_acc@1', acc[0], cnt)
logger.scalar_summary('Spatial/test_acc@5', acc[1], cnt)
logger.scalar_summary('Spatial/test_acc@10', acc[2], cnt)
imgs, labels = pq_train.Get()
pred = model.inference(imgs)
acc = accuracy(pred, labels, topk=(1, 5, 10))
logger.scalar_summary('Spatial/train_acc@1', acc[0], cnt)
logger.scalar_summary('Spatial/train_acc@5', acc[1], cnt)
logger.scalar_summary('Spatial/train_acc@10', acc[2], cnt)
if cnt % 2000 == 0:
savefile = savepath + 'VGG_Spatial_EX1_{:02d}.pt'.format(epoch % 50)
print('Spatial save model to {}'.format(savefile))
torch.save(model.state_dict(), savefile)
if epoch in [10, 20, 50, 60]:
learningrate = learningrate * attenuation
optim = torch.optim.SGD(model.parameters(), lr=learningrate, momentum=0.9, )
def VGG_TwoStream_Video_AVG_Merge_Test():
'''
Video Level VGG TwoStream AVG merge evalution
'''
loops = 100
spa_model = VGG_Spatial_Net().cuda()
tem_model = VGG_Temporal_Net().cuda()
if spa_model_save_file is not None:
spa_model.load_state_dict(torch.load(spa_model_save_file))
print('load spa_model success!')
if tem_model_save_file is not None:
tem_model.load_state_dict(torch.load(tem_model_save_file))
print('load tem_model success!')
imgs, labels = gen()
labels = Variable(torch.from_numpy(labels)).cuda().long()
n, b, c, w, h = imgs.shape
correct_d = [0, 0, 0]
correct_spa = [0, 0, 0]
correct_tmp = [0, 0, 0]
for l in range(loops):
imgs, labels = gen()
print('l: ', l)
print('correct_d', correct_d)
print('correct_spa', correct_spa)
print('correct_tmp', correct_tmp)
for i in range(n):
spatial_input = Variable(torch.from_numpy(imgs[i, :, 0:3, :, :])).cuda().float()
temporal_input = Variable(torch.from_numpy(imgs[i, :, 3:, :, :])).cuda().float()
# print(labels)
# print('spatial_input: ',spatial_input.size())
# print('temporal_input: ',temporal_input.size())
predict_1 = spa_model.inference(spatial_input)
predict_2 = tem_model.inference(temporal_input)
# print('predict_1: ',predict_1.size())
# print('predict_2: ',predict_2.size())
predict_all = (predict_1 + predict_2) / 2
predict_all = torch.cumsum(predict_all, 0)
predict_all = predict_all[-1, :]
predict_all = predict_all / b
predict_all = predict_all.view(1, 101)
target = np.array([labels[i, 0]])
target = torch.from_numpy(target).view(1, 1).cuda().long()
target = Variable(target)
# print('target: ',target)
# print('predict_all: ',predict_all.size())
predict_1 = torch.cumsum(predict_1, 0)
predict_1 = predict_1[-1].view(1, 101)
predict_1 = predict_1 / b
acc = accuracy(predict_1, target, topk=(1, 5, 10))
for ii in range(3):
if acc[ii] > 0.5:
correct_spa[ii] += 1
# print(' only avg spatial net acc:')
# print('acc@1: ',acc[0])
# print('acc@5: ',acc[1])
# print('acc@10: ',acc[2])
# print('-'*20)
#
#
predict_2 = torch.cumsum(predict_2, 0)
predict_2 = predict_2[-1].view(1, 101)
predict_2 = predict_2 / b
acc = accuracy(predict_2, target, topk=(1, 5, 10))
# print(' only avg temporal net acc:')
# print('acc@1: ',acc[0])
# print('acc@5: ',acc[1])
# print('acc@10: ',acc[2])
# print('-'*20)
for ii in range(3):
if acc[ii] > 0.5:
correct_tmp[ii] += 1
acc = accuracy(predict_all, target, topk=(1, 5, 10))
# print(' avg merge two net acc: ')
# print('acc@1: ',acc[0])
# print('acc@5: ',acc[1])
# print('acc@10: ',acc[2])
# print('-'*20)
for ii in range(3):
if acc[ii] > 0.5:
correct_d[ii] += 1
import numpy as np
import torch
from torch.autograd import Variable
from VideoClassification.utils.Others.toolkits import accuracy
out = np.array([[0.1, -0.1, 0], [0.1, -0.2, 0.3], [1, -0.1, 9]])
target = np.array([1, 2, 0])
out = Variable(torch.from_numpy(out))
target = Variable(torch.from_numpy(target))
acc = accuracy(out, target, topk=(1, 2))
acc = [a.data[0] for a in acc]
import torchvision.transforms as transforms
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])