utils/torch_utils.py

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
PyTorch utils
"""

import math
import os
import platform
import subprocess
import time
import warnings
from contextlib import contextmanager
from copy import deepcopy
from pathlib import Path

import torch
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.parallel import DistributedDataParallel as DDP

from utils.general import LOGGER, check_version, colorstr, file_date, git_describe

LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1))  # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))

try:
    import thop  # for FLOPs computation
except ImportError:
    thop = None

# Suppress PyTorch warnings
warnings.filterwarnings('ignore', message='User provided device_type of \'cuda\', but CUDA is not available. Disabling')


def smart_DDP(model):
    # Model DDP creation with checks
    assert not check_version(torch.__version__, '1.12.0', pinned=True), \
        'torch==1.12.0 torchvision==0.13.0 DDP training is not supported due to a known issue. ' \
        'Please upgrade or downgrade torch to use DDP. See https://github.com/ultralytics/yolov5/issues/8395'
    if check_version(torch.__version__, '1.11.0'):
        return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK, static_graph=True)
    else:
        return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK)


@contextmanager
def torch_distributed_zero_first(local_rank: int):
    # Decorator to make all processes in distributed training wait for each local_master to do something
    if local_rank not in [-1, 0]:
        dist.barrier(device_ids=[local_rank])
    yield
    if local_rank == 0:
        dist.barrier(device_ids=[0])


def device_count():
    # Returns number of CUDA devices available. Safe version of torch.cuda.device_count(). Supports Linux and Windows
    assert platform.system() in ('Linux', 'Windows'), 'device_count() only supported on Linux or Windows'
    try:
        cmd = 'nvidia-smi -L | wc -l' if platform.system() == 'Linux' else 'nvidia-smi -L | find /c /v ""'  # Windows
        return int(subprocess.run(cmd, shell=True, capture_output=True, check=True).stdout.decode().split()[-1])
    except Exception:
        return 0


def select_device(device='', batch_size=0, newline=True):
    # device = None or 'cpu' or 0 or '0' or '0,1,2,3'
    s = f'YOLOv5 🚀 {git_describe() or file_date()} Python-{platform.python_version()} torch-{torch.__version__} '
    device = str(device).strip().lower().replace('cuda:', '').replace('none', '')  # to string, 'cuda:0' to '0'
    cpu = device == 'cpu'
    mps = device == 'mps'  # Apple Metal Performance Shaders (MPS)
    if cpu or mps:
        os.environ['CUDA_VISIBLE_DEVICES'] = '-1'  # force torch.cuda.is_available() = False
    elif device:  # non-cpu device requested
        os.environ['CUDA_VISIBLE_DEVICES'] = device  # set environment variable - must be before assert is_available()
        assert torch.cuda.is_available() and torch.cuda.device_count() >= len(device.replace(',', '')), \
            f"Invalid CUDA '--device {device}' requested, use '--device cpu' or pass valid CUDA device(s)"

    if not (cpu or mps) and torch.cuda.is_available():  # prefer GPU if available
        devices = device.split(',') if device else '0'  # range(torch.cuda.device_count())  # i.e. 0,1,6,7
        n = len(devices)  # device count
        if n > 1 and batch_size > 0:  # check batch_size is divisible by device_count
            assert batch_size % n == 0, f'batch-size {batch_size} not multiple of GPU count {n}'
        space = ' ' * (len(s) + 1)
        for i, d in enumerate(devices):
            p = torch.cuda.get_device_properties(i)
            s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / (1 << 20):.0f}MiB)\n"  # bytes to MB
        arg = 'cuda:0'
    elif mps and getattr(torch, 'has_mps', False) and torch.backends.mps.is_available():  # prefer MPS if available
        s += 'MPS\n'
        arg = 'mps'
    else:  # revert to CPU
        s += 'CPU\n'
        arg = 'cpu'

    if not newline:
        s = s.rstrip()
    LOGGER.info(s.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else s)  # emoji-safe
    return torch.device(arg)


def time_sync():
    # PyTorch-accurate time
    if torch.cuda.is_available():
        torch.cuda.synchronize()
    return time.time()


def profile(input, ops, n=10, device=None):
    # YOLOv5 speed/memory/FLOPs profiler
    #
    # Usage:
    #     input = torch.randn(16, 3, 640, 640)
    #     m1 = lambda x: x * torch.sigmoid(x)
    #     m2 = nn.SiLU()
    #     profile(input, [m1, m2], n=100)  # profile over 100 iterations

    results = []
    if not isinstance(device, torch.device):
        device = select_device(device)
    print(f"{'Params':>12s}{'GFLOPs':>12s}{'GPU_mem (GB)':>14s}{'forward (ms)':>14s}{'backward (ms)':>14s}"
          f"{'input':>24s}{'output':>24s}")

    for x in input if isinstance(input, list) else [input]:
        x = x.to(device)
        x.requires_grad = True
        for m in ops if isinstance(ops, list) else [ops]:
            m = m.to(device) if hasattr(m, 'to') else m  # device
            m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m
            tf, tb, t = 0, 0, [0, 0, 0]  # dt forward, backward
            try:
                flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2  # GFLOPs
            except Exception:
                flops = 0

            try:
                for _ in range(n):
                    t[0] = time_sync()
                    y = m(x)
                    t[1] = time_sync()
                    try:
                        _ = (sum(yi.sum() for yi in y) if isinstance(y, list) else y).sum().backward()
                        t[2] = time_sync()
                    except Exception:  # no backward method
                        # print(e)  # for debug
                        t[2] = float('nan')
                    tf += (t[1] - t[0]) * 1000 / n  # ms per op forward
                    tb += (t[2] - t[1]) * 1000 / n  # ms per op backward
                mem = torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0  # (GB)
                s_in, s_out = (tuple(x.shape) if isinstance(x, torch.Tensor) else 'list' for x in (x, y))  # shapes
                p = sum(x.numel() for x in m.parameters()) if isinstance(m, nn.Module) else 0  # parameters
                print(f'{p:12}{flops:12.4g}{mem:>14.3f}{tf:14.4g}{tb:14.4g}{str(s_in):>24s}{str(s_out):>24s}')
                results.append([p, flops, mem, tf, tb, s_in, s_out])
            except Exception as e:
                print(e)
                results.append(None)
            torch.cuda.empty_cache()
    return results


def is_parallel(model):
    # Returns True if model is of type DP or DDP
    return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)


def de_parallel(model):
    # De-parallelize a model: returns single-GPU model if model is of type DP or DDP
    return model.module if is_parallel(model) else model


def initialize_weights(model):
    for m in model.modules():
        t = type(m)
        if t is nn.Conv2d:
            pass  # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
        elif t is nn.BatchNorm2d:
            m.eps = 1e-3
            m.momentum = 0.03
        elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
            m.inplace = True


def find_modules(model, mclass=nn.Conv2d):
    # Finds layer indices matching module class 'mclass'
    return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)]


def sparsity(model):
    # Return global model sparsity
    a, b = 0, 0
    for p in model.parameters():
        a += p.numel()
        b += (p == 0).sum()
    return b / a


def prune(model, amount=0.3):
    # Prune model to requested global sparsity
    import torch.nn.utils.prune as prune
    print('Pruning model... ', end='')
    for name, m in model.named_modules():
        if isinstance(m, nn.Conv2d):
            prune.l1_unstructured(m, name='weight', amount=amount)  # prune
            prune.remove(m, 'weight')  # make permanent
    print(' %.3g global sparsity' % sparsity(model))


def fuse_conv_and_bn(conv, bn):
    # Fuse Conv2d() and BatchNorm2d() layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
    fusedconv = nn.Conv2d(conv.in_channels,
                          conv.out_channels,
                          kernel_size=conv.kernel_size,
                          stride=conv.stride,
                          padding=conv.padding,
                          groups=conv.groups,
                          bias=True).requires_grad_(False).to(conv.weight.device)

    # Prepare filters
    w_conv = conv.weight.clone().view(conv.out_channels, -1)
    w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
    fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))

    # Prepare spatial bias
    b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
    b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
    fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)

    return fusedconv


def model_info(model, verbose=False, img_size=640):
    # Model information. img_size may be int or list, i.e. img_size=640 or img_size=[640, 320]
    n_p = sum(x.numel() for x in model.parameters())  # number parameters
    n_g = sum(x.numel() for x in model.parameters() if x.requires_grad)  # number gradients
    if verbose:
        print(f"{'layer':>5} {'name':>40} {'gradient':>9} {'parameters':>12} {'shape':>20} {'mu':>10} {'sigma':>10}")
        for i, (name, p) in enumerate(model.named_parameters()):
            name = name.replace('module_list.', '')
            print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
                  (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))

    try:  # FLOPs
        from thop import profile
        stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32
        img = torch.zeros((1, model.yaml.get('ch', 3), stride, stride), device=next(model.parameters()).device)  # input
        flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2  # stride GFLOPs
        img_size = img_size if isinstance(img_size, list) else [img_size, img_size]  # expand if int/float
        fs = ', %.1f GFLOPs' % (flops * img_size[0] / stride * img_size[1] / stride)  # 640x640 GFLOPs
    except Exception:
        fs = ''

    name = Path(model.yaml_file).stem.replace('yolov5', 'YOLOv5') if hasattr(model, 'yaml_file') else 'Model'
    LOGGER.info(f"{name} summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")


def scale_img(img, ratio=1.0, same_shape=False, gs=32):  # img(16,3,256,416)
    # Scales img(bs,3,y,x) by ratio constrained to gs-multiple
    if ratio == 1.0:
        return img
    h, w = img.shape[2:]
    s = (int(h * ratio), int(w * ratio))  # new size
    img = F.interpolate(img, size=s, mode='bilinear', align_corners=False)  # resize
    if not same_shape:  # pad/crop img
        h, w = (math.ceil(x * ratio / gs) * gs for x in (h, w))
    return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447)  # value = imagenet mean


def copy_attr(a, b, include=(), exclude=()):
    # Copy attributes from b to a, options to only include [...] and to exclude [...]
    for k, v in b.__dict__.items():
        if (len(include) and k not in include) or k.startswith('_') or k in exclude:
            continue
        else:
            setattr(a, k, v)


def smart_optimizer(model, name='Adam', lr=0.001, momentum=0.9, weight_decay=1e-5):
    # YOLOv5 3-param group optimizer: 0) weights with decay, 1) weights no decay, 2) biases no decay
    g = [], [], []  # optimizer parameter groups
    bn = tuple(v for k, v in nn.__dict__.items() if 'Norm' in k)  # normalization layers, i.e. BatchNorm2d()
    for v in model.modules():
        if hasattr(v, 'bias') and isinstance(v.bias, nn.Parameter):  # bias (no decay)
            g[2].append(v.bias)
        if isinstance(v, bn):  # weight (no decay)
            g[1].append(v.weight)
        elif hasattr(v, 'weight') and isinstance(v.weight, nn.Parameter):  # weight (with decay)
            g[0].append(v.weight)

    if name == 'Adam':
        optimizer = torch.optim.Adam(g[2], lr=lr, betas=(momentum, 0.999))  # adjust beta1 to momentum
    elif name == 'AdamW':
        optimizer = torch.optim.AdamW(g[2], lr=lr, betas=(momentum, 0.999), weight_decay=0.0)
    elif name == 'RMSProp':
        optimizer = torch.optim.RMSprop(g[2], lr=lr, momentum=momentum)
    elif name == 'SGD':
        optimizer = torch.optim.SGD(g[2], lr=lr, momentum=momentum, nesterov=True)
    else:
        raise NotImplementedError(f'Optimizer {name} not implemented.')

    optimizer.add_param_group({'params': g[0], 'weight_decay': weight_decay})  # add g0 with weight_decay
    optimizer.add_param_group({'params': g[1], 'weight_decay': 0.0})  # add g1 (BatchNorm2d weights)
    LOGGER.info(f"{colorstr('optimizer:')} {type(optimizer).__name__} with parameter groups "
                f"{len(g[1])} weight (no decay), {len(g[0])} weight, {len(g[2])} bias")
    return optimizer


class EarlyStopping:
    # YOLOv5 simple early stopper
    def __init__(self, patience=30):
        self.best_fitness = 0.0  # i.e. mAP
        self.best_epoch = 0
        self.patience = patience or float('inf')  # epochs to wait after fitness stops improving to stop
        self.possible_stop = False  # possible stop may occur next epoch

    def __call__(self, epoch, fitness):
        if fitness >= self.best_fitness:  # >= 0 to allow for early zero-fitness stage of training
            self.best_epoch = epoch
            self.best_fitness = fitness
        delta = epoch - self.best_epoch  # epochs without improvement
        self.possible_stop = delta >= (self.patience - 1)  # possible stop may occur next epoch
        stop = delta >= self.patience  # stop training if patience exceeded
        if stop:
            LOGGER.info(f'Stopping training early as no improvement observed in last {self.patience} epochs. '
                        f'Best results observed at epoch {self.best_epoch}, best model saved as best.pt.\n'
                        f'To update EarlyStopping(patience={self.patience}) pass a new patience value, '
                        f'i.e. `python train.py --patience 300` or use `--patience 0` to disable EarlyStopping.')
        return stop


class ModelEMA:
    """ Updated Exponential Moving Average (EMA) from https://github.com/rwightman/pytorch-image-models
    Keeps a moving average of everything in the model state_dict (parameters and buffers)
    For EMA details see https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
    """

    def __init__(self, model, decay=0.9999, tau=2000, updates=0):
        # Create EMA
        self.ema = deepcopy(de_parallel(model)).eval()  # FP32 EMA
        # if next(model.parameters()).device.type != 'cpu':
        #     self.ema.half()  # FP16 EMA
        self.updates = updates  # number of EMA updates
        self.decay = lambda x: decay * (1 - math.exp(-x / tau))  # decay exponential ramp (to help early epochs)
        for p in self.ema.parameters():
            p.requires_grad_(False)

    def update(self, model):
        # Update EMA parameters
        with torch.no_grad():
            self.updates += 1
            d = self.decay(self.updates)

            msd = de_parallel(model).state_dict()  # model state_dict
            for k, v in self.ema.state_dict().items():
                if v.dtype.is_floating_point:
                    v *= d
                    v += (1 - d) * msd[k].detach()

    def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
        # Update EMA attributes
        copy_attr(self.ema, model, include, exclude)