本文程式碼基於 PyTorch 1.0 版本,需要用到以下包
import collections
import os
import shutil
import tqdm
import numpy as np
import PIL.Image
import torch
import torchvision
基礎配置
檢查 PyTorch 版本
torch.__version__ # PyTorch version
torch.version.cuda # Corresponding CUDA version
torch.backends.cudnn.version() # Corresponding cuDNN version
torch.cuda.get_device_name(0) # GPU type
更新 PyTorch
PyTorch 將被安裝在 anaconda3/lib/python3.7/site-packages/torch/目錄下。
conda update pytorch torchvision -c pytorch
固定隨機種子
torch.manual_seed(0)
torch.cuda.manual_seed_all(0)
指定程式執行在特定 GPU 卡上
在命令列指定環境變數
CUDA_VISIBLE_DEVICES=0,1 python train.py
或在程式碼中指定
os.environ['CUDA_VISIBLE_DEVICES'] = '0,1'
判斷是否有 CUDA 支援
torch.cuda.is_available()
設定為 cuDNN benchmark 模式
Benchmark 模式會提升計算速度,但是由於計算中有隨機性,每次網路前饋結果略有差異。
torch.backends.cudnn.benchmark = True
如果想要避免這種結果波動,設定
torch.backends.cudnn.deterministic = True
清除 GPU 儲存
有時 Control-C 中止執行後 GPU 儲存沒有及時釋放,需要手動清空。在 PyTorch 內部可以
torch.cuda.empty_cache()
或在命令列可以先使用 ps 找到程式的 PID,再使用 kill 結束該程序
ps aux | grep pythonkill -9 [pid]
或者直接重置沒有被清空的 GPU
nvidia-smi --gpu-reset -i [gpu_id]
張量處理
張量基本資訊
tensor.type() # Data type
tensor.size() # Shape of the tensor. It is a subclass of Python tuple
tensor.dim() # Number of dimensions.
資料型別轉換
# Set default tensor type. Float in PyTorch is much faster than double.
torch.set_default_tensor_type(torch.FloatTensor)
# Type convertions.
tensor = tensor.cuda()
tensor = tensor.cpu()
tensor = tensor.float()
tensor = tensor.long()
torch.Tensor 與 np.ndarray 轉換
# torch.Tensor -> np.ndarray.
ndarray = tensor.cpu().numpy()
# np.ndarray -> torch.Tensor.
tensor = torch.from_numpy(ndarray).float()
tensor = torch.from_numpy(ndarray.copy()).float() # If ndarray has negative stride
torch.Tensor 與 PIL.Image 轉換
PyTorch 中的張量預設採用 N×D×H×W 的順序,並且資料範圍在 [0, 1],需要進行轉置和規範化。
# torch.Tensor -> PIL.Image.
image = PIL.Image.fromarray(torch.clamp(tensor * 255, min=0, max=255
).byte().permute(1, 2, 0).cpu().numpy())
image = torchvision.transforms.functional.to_pil_image(tensor) # Equivalently way
# PIL.Image -> torch.Tensor.
tensor = torch.from_numpy(np.asarray(PIL.Image.open(path))
).permute(2, 0, 1).float() / 255
tensor = torchvision.transforms.functional.to_tensor(PIL.Image.open(path)) # Equivalently way
np.ndarray 與 PIL.Image 轉換
# np.ndarray -> PIL.Image.
image = PIL.Image.fromarray(ndarray.astypde(np.uint8))
# PIL.Image -> np.ndarray.
ndarray = np.asarray(PIL.Image.open(path))
從只包含一個元素的張量中提取值
這在訓練時統計 loss 的變化過程中特別有用。否則這將累積計算圖,使 GPU 儲存佔用量越來越大。
value = tensor.item()
張量形變
張量形變常常需要用於將卷積層特徵輸入全連線層的情形。相比 torch.view,torch.reshape 可以自動處理輸入張量不連續的情況。
tensor = torch.reshape(tensor, shape)
打亂順序
tensor = tensor[torch.randperm(tensor.size(0))] # Shuffle the first dimension
水平翻轉
PyTorch 不支援 tensor[::-1] 這樣的負步長操作,水平翻轉可以用張量索引實現。
# Assume tensor has shape N*D*H*W.tensor = tensor[:, :, :, torch.arange(tensor.size(3) - 1, -1, -1).long()]
複製張量
有三種複製的方式,對應不同的需求。
# Operation | New/Shared memory | Still in computation graph |
tensor.clone() # | New | Yes |
tensor.detach() # | Shared | No |
tensor.detach.clone()() # | New | No |
拼接張量
注意 torch.cat 和 torch.stack 的區別在於 torch.cat 沿著給定的維度拼接,而 torch.stack 會新增一維。例如當引數是 3 個 10×5 的張量,torch.cat 的結果是 30×5 的張量,而 torch.stack 的結果是 3×10×5 的張量。
tensor = torch.cat(list_of_tensors, dim=0)
tensor = torch.stack(list_of_tensors, dim=0)
將整數標記轉換成獨熱(one-hot)編碼
PyTorch 中的標記預設從 0 開始。
N = tensor.size(0)
one_hot = torch.zeros(N, num_classes).long()
one_hot.scatter_(dim=1, index=torch.unsqueeze(tensor, dim=1), src=torch.ones(N, num_classes).long())
得到非零/零元素
torch.nonzero(tensor) # Index of non-zero elements
torch.nonzero(tensor == 0) # Index of zero elements
torch.nonzero(tensor).size(0) # Number of non-zero elements
torch.nonzero(tensor == 0).size(0) # Number of zero elements
張量擴充套件
# Expand tensor of shape 64*512 to shape 64*512*7*7.
torch.reshape(tensor, (64, 512, 1, 1)).expand(64, 512, 7, 7)
矩陣乘法
# Matrix multiplication: (m*n) * (n*p) -> (m*p).
result = torch.mm(tensor1, tensor2)
# Batch matrix multiplication: (b*m*n) * (b*n*p) -> (b*m*p).
result = torch.bmm(tensor1, tensor2)
# Element-wise multiplication.
result = tensor1 * tensor2
計算兩組資料之間的兩兩歐式距離
# X1 is of shape m*d.
X1 = torch.unsqueeze(X1, dim=1).expand(m, n, d)
# X2 is of shape n*d.
X2 = torch.unsqueeze(X2, dim=0).expand(m, n, d)
# dist is of shape m*n, where dist[i][j] = sqrt(|X1[i, :] - X[j, :]|^2)
dist = torch.sqrt(torch.sum((X1 - X2) ** 2, dim=2))
模型定義
卷積層
最常用的卷積層配置是
conv = torch.nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding=1, bias=True)conv = torch.nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0, bias=True)
如果卷積層配置比較複雜,不方便計算輸出大小時,可以利用如下視覺化工具輔助
連結:https://ezyang.github.io/convolution-visualizer/index.html
0GAP(Global average pooling)層
gap = torch.nn.AdaptiveAvgPool2d(output_size=1)
雙線性匯合(bilinear pooling)
X = torch.reshape(N, D, H * W) # Assume X has shape N*D*H*W
X = torch.bmm(X, torch.transpose(X, 1, 2)) / (H * W) # Bilinear pooling
assert X.size() == (N, D, D)
X = torch.reshape(X, (N, D * D))
X = torch.sign(X) * torch.sqrt(torch.abs(X) + 1e-5) # Signed-sqrt normalization
X = torch.nn.functional.normalize(X) # L2 normalization
多卡同步 BN(Batch normalization)
當使用 torch.nn.DataParallel 將程式碼執行在多張 GPU 卡上時,PyTorch 的 BN 層預設操作是各卡上資料獨立地計算均值和標準差,同步 BN 使用所有卡上的資料一起計算 BN 層的均值和標準差,緩解了當批次大小(batch size)比較小時對均值和標準差估計不準的情況,是在目標檢測等任務中一個有效的提升效能的技巧。
連結:https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
類似 BN 滑動平均
如果要實現類似 BN 滑動平均的操作,在 forward 函式中要使用原地(inplace)操作給滑動平均賦值。
class BN(torch.nn.Module)
def __init__(self):
...
self.register_buffer('running_mean', torch.zeros(num_features))
def forward(self, X):
...
self.running_mean += momentum * (current - self.running_mean)
計算模型整體引數量
num_parameters = sum(torch.numel(parameter) for parameter in model.parameters())
類似 Keras 的 model.summary() 輸出模型資訊
連結:https://github.com/sksq96/pytorch-summary
模型權值初始化
注意 model.modules() 和 model.children() 的區別:model.modules() 會迭代地遍歷模型的所有子層,而 model.children() 只會遍歷模型下的一層。
# Common practise for initialization.
for layer in model.modules():
if isinstance(layer, torch.nn.Conv2d):
torch.nn.init.kaiming_normal_(layer.weight, mode='fan_out',
nonlinearity='relu')
if layer.bias is not None:
torch.nn.init.constant_(layer.bias, val=0.0)
elif isinstance(layer, torch.nn.BatchNorm2d):
torch.nn.init.constant_(layer.weight, val=1.0)
torch.nn.init.constant_(layer.bias, val=0.0)
elif isinstance(layer, torch.nn.Linear):
torch.nn.init.xavier_normal_(layer.weight)
if layer.bias is not None:
torch.nn.init.constant_(layer.bias, val=0.0)
# Initialization with given tensor.
layer.weight = torch.nn.Parameter(tensor)
部分層使用預訓練模型
注意如果儲存的模型是 torch.nn.DataParallel,則當前的模型也需要是
model.load_state_dict(torch.load('model,pth'), strict=False)
將在 GPU 儲存的模型載入到 CPU
model.load_state_dict(torch.load('model,pth', map_location='cpu'))
資料準備、特徵提取與微調
得到影片資料基本資訊
import cv2
video = cv2.VideoCapture(mp4_path)
height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT))
width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH))
num_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
fps = int(video.get(cv2.CAP_PROP_FPS))
video.release()
TSN 每段(segment)取樣一幀影片
K = self._num_segments
if is_train:
if num_frames > K:
# Random index for each segment.
frame_indices = torch.randint(
high=num_frames // K, size=(K,), dtype=torch.long)
frame_indices += num_frames // K * torch.arange(K)
else:
frame_indices = torch.randint(
high=num_frames, size=(K - num_frames,), dtype=torch.long)
frame_indices = torch.sort(torch.cat((
torch.arange(num_frames), frame_indices)))[0]
else:
if num_frames > K:
# Middle index for each segment.
frame_indices = num_frames / K // 2
frame_indices += num_frames // K * torch.arange(K)
else:
frame_indices = torch.sort(torch.cat((
torch.arange(num_frames), torch.arange(K - num_frames))))[0]
assert frame_indices.size() == (K,)
return [frame_indices[i] for i in range(K)]
提取 ImageNet 預訓練模型某層的卷積特徵
# VGG-16 relu5-3 feature.
model = torchvision.models.vgg16(pretrained=True).features[:-1]
# VGG-16 pool5 feature.
model = torchvision.models.vgg16(pretrained=True).features
# VGG-16 fc7 feature.
model = torchvision.models.vgg16(pretrained=True)
model.classifier = torch.nn.Sequential(*list(model.classifier.children())[:-3])
# ResNet GAP feature.
model = torchvision.models.resnet18(pretrained=True)
model = torch.nn.Sequential(collections.OrderedDict(
list(model.named_children())[:-1]))
with torch.no_grad():
model.eval()
conv_representation = model(image)
提取 ImageNet 預訓練模型多層的卷積特徵
class FeatureExtractor(torch.nn.Module):
"""Helper class to extract several convolution features from the given
pre-trained model.
Attributes:
_model, torch.nn.Module.
_layers_to_extract, list<str> or set<str>
Example:
>>> model = torchvision.models.resnet152(pretrained=True)
>>> model = torch.nn.Sequential(collections.OrderedDict(
list(model.named_children())[:-1]))
>>> conv_representation = FeatureExtractor(
pretrained_model=model,
layers_to_extract={'layer1', 'layer2', 'layer3', 'layer4'})(image)
"""
def __init__(self, pretrained_model, layers_to_extract):
torch.nn.Module.__init__(self)
self._model = pretrained_model
self._model.eval()
self._layers_to_extract = set(layers_to_extract)
def forward(self, x):
with torch.no_grad():
conv_representation = []
for name, layer in self._model.named_children():
x = layer(x)
if name in self._layers_to_extract:
conv_representation.append(x)
return conv_representation
其他預訓練模型
連結:https://github.com/Cadene/pretrained-models.pytorch
微調全連線層
model = torchvision.models.resnet18(pretrained=True)
for param in model.parameters():
param.requires_grad = False
model.fc = nn.Linear(512, 100) # Replace the last fc layer
optimizer = torch.optim.SGD(model.fc.parameters(), lr=1e-2, momentum=0.9, weight_decay=1e-4)
以較大學習率微調全連線層,較小學習率微調卷積層
model = torchvision.models.resnet18(pretrained=True)
finetuned_parameters = list(map(id, model.fc.parameters()))
conv_parameters = (p for p in model.parameters() if id(p) not in finetuned_parameters)
parameters = [{'params': conv_parameters, 'lr': 1e-3},
{'params': model.fc.parameters()}]
optimizer = torch.optim.SGD(parameters, lr=1e-2, momentum=0.9, weight_decay=1e-4)
模型訓練
常用訓練和驗證資料預處理
其中 ToTensor 操作會將 PIL.Image 或形狀為 H×W×D,數值範圍為 [0, 255] 的 np.ndarray 轉換為形狀為 D×H×W,數值範圍為 [0.0, 1.0] 的 torch.Tensor。
train_transform = torchvision.transforms.Compose([
torchvision.transforms.RandomResizedCrop(size=224,
scale=(0.08, 1.0)),
torchvision.transforms.RandomHorizontalFlip(),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225)),
])
val_transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225)),
])
訓練基本程式碼框架
for t in epoch(80):
for images, labels in tqdm.tqdm(train_loader, desc='Epoch %3d' % (t + 1)):
images, labels = images.cuda(), labels.cuda()
scores = model(images)
loss = loss_function(scores, labels)
optimizer.zero_grad()
loss.backward()
optimizer.step()
標記平滑(label smoothing)
for images, labels in train_loader:
images, labels = images.cuda(), labels.cuda()
N = labels.size(0)
# C is the number of classes.
smoothed_labels = torch.full(size=(N, C), fill_value=0.1 / (C - 1)).cuda()
smoothed_labels.scatter_(dim=1, index=torch.unsqueeze(labels, dim=1), value=0.9)
score = model(images)
log_prob = torch.nn.functional.log_softmax(score, dim=1)
loss = -torch.sum(log_prob * smoothed_labels) / N
optimizer.zero_grad()
loss.backward()
optimizer.step()
Mixup
beta_distribution = torch.distributions.beta.Beta(alpha, alpha)
for images, labels in train_loader:
images, labels = images.cuda(), labels.cuda()
# Mixup images.
lambda_ = beta_distribution.sample([]).item()
index = torch.randperm(images.size(0)).cuda()
mixed_images = lambda_ * images + (1 - lambda_) * images[index, :]
# Mixup loss.
scores = model(mixed_images)
loss = (lambda_ * loss_function(scores, labels)
+ (1 - lambda_) * loss_function(scores, labels[index]))
optimizer.zero_grad()
loss.backward()
optimizer.step()
L1 正則化
l1_regularization = torch.nn.L1Loss(reduction='sum')
loss = ... # Standard cross-entropy loss
for param in model.parameters():
loss += torch.sum(torch.abs(param))
loss.backward()
不對偏置項進行 L2 正則化/權值衰減(weight decay)
bias_list = (param for name, param in model.named_parameters() if name[-4:] == 'bias')
others_list = (param for name, param in model.named_parameters() if name[-4:] != 'bias')
parameters = [{'parameters': bias_list, 'weight_decay': 0},
{'parameters': others_list}]
optimizer = torch.optim.SGD(parameters, lr=1e-2, momentum=0.9, weight_decay=1e-4)
梯度裁剪(gradient clipping)
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=20)
計算 Softmax 輸出的準確率
score = model(images)
prediction = torch.argmax(score, dim=1)
num_correct = torch.sum(prediction == labels).item()
accuruacy = num_correct / labels.size(0)
視覺化模型前饋的計算圖
連結:https://github.com/szagoruyko/pytorchviz
視覺化學習曲線
有 Facebook 自己開發的 Visdom 和 Tensorboard 兩個選擇。
https://github.com/facebookresearch/visdom
https://github.com/lanpa/tensorboardX
# Example using Visdom.
vis = visdom.Visdom(env='Learning curve', use_incoming_socket=False)
assert self._visdom.check_connection()
self._visdom.close()
options = collections.namedtuple('Options', ['loss', 'acc', 'lr'])(
loss={'xlabel': 'Epoch', 'ylabel': 'Loss', 'showlegend': True},
acc={'xlabel': 'Epoch', 'ylabel': 'Accuracy', 'showlegend': True},
lr={'xlabel': 'Epoch', 'ylabel': 'Learning rate', 'showlegend': True})
for t in epoch(80):
tran(...)
val(...)
vis.line(X=torch.Tensor([t + 1]), Y=torch.Tensor([train_loss]),
name='train', win='Loss', update='append', opts=options.loss)
vis.line(X=torch.Tensor([t + 1]), Y=torch.Tensor([val_loss]),
name='val', win='Loss', update='append', opts=options.loss)
vis.line(X=torch.Tensor([t + 1]), Y=torch.Tensor([train_acc]),
name='train', win='Accuracy', update='append', opts=options.acc)
vis.line(X=torch.Tensor([t + 1]), Y=torch.Tensor([val_acc]),
name='val', win='Accuracy', update='append', opts=options.acc)
vis.line(X=torch.Tensor([t + 1]), Y=torch.Tensor([lr]),
win='Learning rate', update='append', opts=options.lr)
得到當前學習率
# If there is one global learning rate (which is the common case).
lr = next(iter(optimizer.param_groups))['lr']
# If there are multiple learning rates for different layers.
all_lr = []
for param_group in optimizer.param_groups:
all_lr.append(param_group['lr'])
學習率衰減
# Reduce learning rate when validation accuarcy plateau.
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='max', patience=5, verbose=True)
for t in range(0, 80):
train(...); val(...)
scheduler.step(val_acc)
# Cosine annealing learning rate.
scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=80)
# Reduce learning rate by 10 at given epochs.
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[50, 70], gamma=0.1)
for t in range(0, 80):
scheduler.step()
train(...); val(...)
# Learning rate warmup by 10 epochs.
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda t: t / 10)
for t in range(0, 10):
scheduler.step()
train(...); val(...)
儲存與載入斷點
注意為了能夠恢復訓練,我們需要同時儲存模型和最佳化器的狀態,以及當前的訓練輪數。
# Save checkpoint.
is_best = current_acc > best_acc
best_acc = max(best_acc, current_acc)
checkpoint = {
'best_acc': best_acc,
'epoch': t + 1,
'model': model.state_dict(),
'optimizer': optimizer.state_dict(),
}
model_path = os.path.join('model', 'checkpoint.pth.tar')
torch.save(checkpoint, model_path)
if is_best:
shutil.copy('checkpoint.pth.tar', model_path)
# Load checkpoint.
if resume:
model_path = os.path.join('model', 'checkpoint.pth.tar')
assert os.path.isfile(model_path)
checkpoint = torch.load(model_path)
best_acc = checkpoint['best_acc']
start_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['model'])
optimizer.load_state_dict(checkpoint['optimizer'])
print('Load checkpoint at epoch %d.' % start_epoch)
計算準確率、查準率(precision)、查全率(recall)
# data['label'] and data['prediction'] are groundtruth label and prediction
# for each image, respectively.
accuracy = np.mean(data['label'] == data['prediction']) * 100
# Compute recision and recall for each class.
for c in range(len(num_classes)):
tp = np.dot((data['label'] == c).astype(int),
(data['prediction'] == c).astype(int))
tp_fp = np.sum(data['prediction'] == c)
tp_fn = np.sum(data['label'] == c)
precision = tp / tp_fp * 100
recall = tp / tp_fn * 100
PyTorch 其他注意事項
模型定義
建議有引數的層和匯合(pooling)層使用 torch.nn 模組定義,啟用函式直接使用 torch.nn.functional。torch.nn 模組和 torch.nn.functional 的區別在於,torch.nn 模組在計算時底層呼叫了 torch.nn.functional,但 torch.nn 模組包括該層引數,還可以應對訓練和測試兩種網路狀態。使用 torch.nn.functional 時要注意網路狀態,如
def forward(self, x):
...
x = torch.nn.functional.dropout(x, p=0.5, training=self.training)
model(x) 前用 model.train() 和 model.eval() 切換網路狀態。
不需要計算梯度的程式碼塊用 with torch.no_grad() 包含起來。model.eval() 和 torch.no_grad() 的區別在於,model.eval() 是將網路切換為測試狀態,例如 BN 和隨機失活(dropout)在訓練和測試階段使用不同的計算方法。torch.no_grad() 是關閉 PyTorch 張量的自動求導機制,以減少儲存使用和加速計算,得到的結果無法進行 loss.backward()。
torch.nn.CrossEntropyLoss 的輸入不需要經過 Softmax。torch.nn.CrossEntropyLoss 等價於 torch.nn.functional.log_softmax + torch.nn.NLLLoss。
loss.backward() 前用 optimizer.zero_grad() 清除累積梯度。optimizer.zero_grad() 和 model.zero_grad() 效果一樣。
PyTorch 效能與除錯
torch.utils.data.DataLoader 中儘量設定 pin_memory=True,對特別小的資料集如 MNIST 設定 pin_memory=False 反而更快一些。num_workers 的設定需要在實驗中找到最快的取值。
用 del 及時刪除不用的中間變數,節約 GPU 儲存。
使用 inplace 操作可節約 GPU 儲存,如
x = torch.nn.functional.relu(x, inplace=True)
減少 CPU 和 GPU 之間的資料傳輸。例如如果你想知道一個 epoch 中每個 mini-batch 的 loss 和準確率,先將它們累積在 GPU 中等一個 epoch 結束之後一起傳輸回 CPU 會比每個 mini-batch 都進行一次 GPU 到 CPU 的傳輸更快。
使用半精度浮點數 half() 會有一定的速度提升,具體效率依賴於 GPU 型號。需要小心數值精度過低帶來的穩定性問題。
時常使用 assert tensor.size() == (N, D, H, W) 作為除錯手段,確保張量維度和你設想中一致。
除了標記 y 外,儘量少使用一維張量,使用 n*1 的二維張量代替,可以避免一些意想不到的一維張量計算結果。
統計程式碼各部分耗時
with torch.autograd.profiler.profile(enabled=True, use_cuda=False) as profile:
...
print(profile)
或者在命令列執行
python -m torch.utils.bottleneck main.py
致謝
感謝 @些許流年和@El tnoto的勘誤。由於作者才疏學淺,更兼時間和精力所限,程式碼中錯誤之處在所難免,敬請讀者批評指正。
參考資料
PyTorch 官方程式碼:pytorch/examples (https://link.zhihu.com/?target=https%3A//github.com/pytorch/examples)
PyTorch 論壇:PyTorch Forums (https://link.zhihu.com/?target=https%3A//discuss.pytorch.org/latest%3Forder%3Dviews)
PyTorch 文件:http://pytorch.org/docs/stable/index.html (https://link.zhihu.com/?target=http%3A//pytorch.org/docs/stable/index.html)
其他基於 PyTorch 的公開實現程式碼,無法一一列舉
張皓:南京大學計算機系機器學習與資料探勘所(LAMDA)碩士生,研究方向為計算機視覺和機器學習,特別是視覺識別和深度學習。個人主頁:http://lamda.nju.edu.cn/zhangh/
原知乎連結:https://zhuanlan.zhihu.com/p/59205847?