|
|
|
""" |
|
Train a YOLOv5 model on a custom dataset |
|
|
|
Usage: |
|
$ python path/to/train.py --data coco128.yaml --weights yolov5s.pt --img 640 |
|
""" |
|
|
|
import argparse |
|
import logging |
|
import math |
|
import os |
|
import random |
|
import sys |
|
import time |
|
from copy import deepcopy |
|
from pathlib import Path |
|
|
|
import numpy as np |
|
import torch |
|
import torch.distributed as dist |
|
import torch.nn as nn |
|
import yaml |
|
from torch.cuda import amp |
|
from torch.nn.parallel import DistributedDataParallel as DDP |
|
from torch.optim import SGD, Adam, lr_scheduler |
|
from tqdm import tqdm |
|
|
|
FILE = Path(__file__).absolute() |
|
sys.path.append(FILE.parents[0].as_posix()) |
|
|
|
import val |
|
from models.experimental import attempt_load |
|
from models.yolo import Model |
|
from utils.autoanchor import check_anchors |
|
from utils.callbacks import Callbacks |
|
from utils.datasets import create_dataloader |
|
from utils.downloads import attempt_download |
|
from utils.general import ( |
|
check_dataset, |
|
check_file, |
|
check_git_status, |
|
check_img_size, |
|
check_requirements, |
|
check_suffix, |
|
check_yaml, |
|
colorstr, |
|
get_latest_run, |
|
increment_path, |
|
init_seeds, |
|
labels_to_class_weights, |
|
labels_to_image_weights, |
|
methods, |
|
one_cycle, |
|
print_mutation, |
|
set_logging, |
|
strip_optimizer, |
|
) |
|
from utils.loggers import Loggers |
|
from utils.loggers.wandb.wandb_utils import check_wandb_resume |
|
from utils.loss import ComputeLoss |
|
from utils.metrics import fitness |
|
from utils.plots import plot_evolve, plot_labels |
|
from utils.torch_utils import ( |
|
EarlyStopping, |
|
ModelEMA, |
|
de_parallel, |
|
intersect_dicts, |
|
select_device, |
|
torch_distributed_zero_first, |
|
) |
|
|
|
LOGGER = logging.getLogger(__name__) |
|
LOCAL_RANK = int( |
|
os.getenv("LOCAL_RANK", -1) |
|
) |
|
RANK = int(os.getenv("RANK", -1)) |
|
WORLD_SIZE = int(os.getenv("WORLD_SIZE", 1)) |
|
|
|
|
|
def train(hyp, opt, device, callbacks): |
|
( |
|
save_dir, |
|
epochs, |
|
batch_size, |
|
weights, |
|
single_cls, |
|
evolve, |
|
data, |
|
cfg, |
|
resume, |
|
noval, |
|
nosave, |
|
workers, |
|
freeze, |
|
) = ( |
|
Path(opt.save_dir), |
|
opt.epochs, |
|
opt.batch_size, |
|
opt.weights, |
|
opt.single_cls, |
|
opt.evolve, |
|
opt.data, |
|
opt.cfg, |
|
opt.resume, |
|
opt.noval, |
|
opt.nosave, |
|
opt.workers, |
|
opt.freeze, |
|
) |
|
|
|
|
|
w = save_dir / "weights" |
|
w.mkdir(parents=True, exist_ok=True) |
|
last, best = w / "last.pt", w / "best.pt" |
|
|
|
|
|
if isinstance(hyp, str): |
|
with open(hyp) as f: |
|
hyp = yaml.safe_load(f) |
|
LOGGER.info( |
|
colorstr("hyperparameters: ") |
|
+ ", ".join(f"{k}={v}" for k, v in hyp.items()) |
|
) |
|
|
|
|
|
with open(save_dir / "hyp.yaml", "w") as f: |
|
yaml.safe_dump(hyp, f, sort_keys=False) |
|
with open(save_dir / "opt.yaml", "w") as f: |
|
yaml.safe_dump(vars(opt), f, sort_keys=False) |
|
data_dict = None |
|
|
|
|
|
if RANK in [-1, 0]: |
|
loggers = Loggers( |
|
save_dir, weights, opt, hyp, LOGGER |
|
) |
|
if loggers.wandb: |
|
data_dict = loggers.wandb.data_dict |
|
if resume: |
|
weights, epochs, hyp = opt.weights, opt.epochs, opt.hyp |
|
|
|
|
|
for k in methods(loggers): |
|
callbacks.register_action(k, callback=getattr(loggers, k)) |
|
|
|
|
|
plots = not evolve |
|
cuda = device.type != "cpu" |
|
init_seeds(1 + RANK) |
|
with torch_distributed_zero_first(RANK): |
|
data_dict = data_dict or check_dataset(data) |
|
train_path, val_path = data_dict["train"], data_dict["val"] |
|
nc = 1 if single_cls else int(data_dict["nc"]) |
|
names = ( |
|
["item"] |
|
if single_cls and len(data_dict["names"]) != 1 |
|
else data_dict["names"] |
|
) |
|
assert ( |
|
len(names) == nc |
|
), f"{len(names)} names found for nc={nc} dataset in {data}" |
|
is_coco = data.endswith("coco.yaml") and nc == 80 |
|
|
|
|
|
check_suffix(weights, ".pt") |
|
pretrained = weights.endswith(".pt") |
|
if pretrained: |
|
with torch_distributed_zero_first(RANK): |
|
weights = attempt_download( |
|
weights |
|
) |
|
ckpt = torch.load(weights, map_location=device) |
|
model = Model( |
|
cfg or ckpt["model"].yaml, ch=3, nc=nc, anchors=hyp.get("anchors") |
|
).to( |
|
device |
|
) |
|
exclude = ( |
|
["anchor"] if (cfg or hyp.get("anchors")) and not resume else [] |
|
) |
|
csd = ( |
|
ckpt["model"].float().state_dict() |
|
) |
|
csd = intersect_dicts( |
|
csd, model.state_dict(), exclude=exclude |
|
) |
|
model.load_state_dict(csd, strict=False) |
|
LOGGER.info( |
|
f"Transferred {len(csd)}/{len(model.state_dict())} items from {weights}" |
|
) |
|
else: |
|
model = Model(cfg, ch=3, nc=nc, anchors=hyp.get("anchors")).to( |
|
device |
|
) |
|
|
|
|
|
freeze = [f"model.{x}." for x in range(freeze)] |
|
for k, v in model.named_parameters(): |
|
v.requires_grad = True |
|
if any(x in k for x in freeze): |
|
print(f"freezing {k}") |
|
v.requires_grad = False |
|
|
|
|
|
nbs = 64 |
|
accumulate = max( |
|
round(nbs / batch_size), 1 |
|
) |
|
hyp["weight_decay"] *= batch_size * accumulate / nbs |
|
LOGGER.info(f"Scaled weight_decay = {hyp['weight_decay']}") |
|
|
|
g0, g1, g2 = [], [], [] |
|
for v in model.modules(): |
|
if hasattr(v, "bias") and isinstance(v.bias, nn.Parameter): |
|
g2.append(v.bias) |
|
if isinstance(v, nn.BatchNorm2d): |
|
g0.append(v.weight) |
|
elif hasattr(v, "weight") and isinstance( |
|
v.weight, nn.Parameter |
|
): |
|
g1.append(v.weight) |
|
|
|
if opt.adam: |
|
optimizer = Adam( |
|
g0, lr=hyp["lr0"], betas=(hyp["momentum"], 0.999) |
|
) |
|
else: |
|
optimizer = SGD( |
|
g0, lr=hyp["lr0"], momentum=hyp["momentum"], nesterov=True |
|
) |
|
|
|
optimizer.add_param_group( |
|
{"params": g1, "weight_decay": hyp["weight_decay"]} |
|
) |
|
optimizer.add_param_group({"params": g2}) |
|
LOGGER.info( |
|
f"{colorstr('optimizer:')} {type(optimizer).__name__} with parameter groups " |
|
f"{len(g0)} weight, {len(g1)} weight (no decay), {len(g2)} bias" |
|
) |
|
del g0, g1, g2 |
|
|
|
|
|
if opt.linear_lr: |
|
lf = ( |
|
lambda x: (1 - x / (epochs - 1)) * (1.0 - hyp["lrf"]) + hyp["lrf"] |
|
) |
|
else: |
|
lf = one_cycle(1, hyp["lrf"], epochs) |
|
scheduler = lr_scheduler.LambdaLR( |
|
optimizer, lr_lambda=lf |
|
) |
|
|
|
|
|
ema = ModelEMA(model) if RANK in [-1, 0] else None |
|
|
|
|
|
start_epoch, best_fitness = 0, 0.0 |
|
if pretrained: |
|
|
|
if ckpt["optimizer"] is not None: |
|
optimizer.load_state_dict(ckpt["optimizer"]) |
|
best_fitness = ckpt["best_fitness"] |
|
|
|
|
|
if ema and ckpt.get("ema"): |
|
ema.ema.load_state_dict(ckpt["ema"].float().state_dict()) |
|
ema.updates = ckpt["updates"] |
|
|
|
|
|
start_epoch = ckpt["epoch"] + 1 |
|
if resume: |
|
assert ( |
|
start_epoch > 0 |
|
), f"{weights} training to {epochs} epochs is finished, nothing to resume." |
|
if epochs < start_epoch: |
|
LOGGER.info( |
|
f"{weights} has been trained for {ckpt['epoch']} epochs. Fine-tuning for {epochs} more epochs." |
|
) |
|
epochs += ckpt["epoch"] |
|
|
|
del ckpt, csd |
|
|
|
|
|
gs = max(int(model.stride.max()), 32) |
|
nl = model.model[ |
|
-1 |
|
].nl |
|
imgsz = check_img_size( |
|
opt.imgsz, gs, floor=gs * 2 |
|
) |
|
|
|
|
|
if cuda and RANK == -1 and torch.cuda.device_count() > 1: |
|
logging.warning( |
|
"DP not recommended, instead use torch.distributed.run for best DDP Multi-GPU results.\n" |
|
"See Multi-GPU Tutorial at https://github.com/ultralytics/yolov5/issues/475 to get started." |
|
) |
|
model = torch.nn.DataParallel(model) |
|
|
|
|
|
if opt.sync_bn and cuda and RANK != -1: |
|
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device) |
|
LOGGER.info("Using SyncBatchNorm()") |
|
|
|
|
|
train_loader, dataset = create_dataloader( |
|
train_path, |
|
imgsz, |
|
batch_size // WORLD_SIZE, |
|
gs, |
|
single_cls, |
|
hyp=hyp, |
|
augment=True, |
|
cache=opt.cache, |
|
rect=opt.rect, |
|
rank=RANK, |
|
workers=workers, |
|
image_weights=opt.image_weights, |
|
quad=opt.quad, |
|
prefix=colorstr("train: "), |
|
) |
|
mlc = int(np.concatenate(dataset.labels, 0)[:, 0].max()) |
|
nb = len(train_loader) |
|
assert ( |
|
mlc < nc |
|
), f"Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}" |
|
|
|
|
|
if RANK in [-1, 0]: |
|
val_loader = create_dataloader( |
|
val_path, |
|
imgsz, |
|
batch_size // WORLD_SIZE * 2, |
|
gs, |
|
single_cls, |
|
hyp=hyp, |
|
cache=None if noval else opt.cache, |
|
rect=True, |
|
rank=-1, |
|
workers=workers, |
|
pad=0.5, |
|
prefix=colorstr("val: "), |
|
)[0] |
|
|
|
if not resume: |
|
labels = np.concatenate(dataset.labels, 0) |
|
|
|
|
|
|
|
if plots: |
|
plot_labels(labels, names, save_dir) |
|
|
|
|
|
if not opt.noautoanchor: |
|
check_anchors( |
|
dataset, model=model, thr=hyp["anchor_t"], imgsz=imgsz |
|
) |
|
model.half().float() |
|
|
|
callbacks.run("on_pretrain_routine_end") |
|
|
|
|
|
if cuda and RANK != -1: |
|
model = DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK) |
|
|
|
|
|
hyp["box"] *= 3.0 / nl |
|
hyp["cls"] *= nc / 80.0 * 3.0 / nl |
|
hyp["obj"] *= ( |
|
(imgsz / 640) ** 2 * 3.0 / nl |
|
) |
|
hyp["label_smoothing"] = opt.label_smoothing |
|
model.nc = nc |
|
model.hyp = hyp |
|
model.class_weights = ( |
|
labels_to_class_weights(dataset.labels, nc).to(device) * nc |
|
) |
|
model.names = names |
|
|
|
|
|
t0 = time.time() |
|
nw = max( |
|
round(hyp["warmup_epochs"] * nb), 1000 |
|
) |
|
|
|
last_opt_step = -1 |
|
maps = np.zeros(nc) |
|
results = ( |
|
0, |
|
0, |
|
0, |
|
0, |
|
0, |
|
0, |
|
0, |
|
) |
|
scheduler.last_epoch = start_epoch - 1 |
|
scaler = amp.GradScaler(enabled=cuda) |
|
stopper = EarlyStopping(patience=opt.patience) |
|
compute_loss = ComputeLoss(model) |
|
LOGGER.info( |
|
f"Image sizes {imgsz} train, {imgsz} val\n" |
|
f"Using {train_loader.num_workers} dataloader workers\n" |
|
f"Logging results to {colorstr('bold', save_dir)}\n" |
|
f"Starting training for {epochs} epochs..." |
|
) |
|
for epoch in range( |
|
start_epoch, epochs |
|
): |
|
model.train() |
|
|
|
|
|
if opt.image_weights: |
|
cw = ( |
|
model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc |
|
) |
|
iw = labels_to_image_weights( |
|
dataset.labels, nc=nc, class_weights=cw |
|
) |
|
dataset.indices = random.choices( |
|
range(dataset.n), weights=iw, k=dataset.n |
|
) |
|
|
|
|
|
|
|
|
|
|
|
mloss = torch.zeros(3, device=device) |
|
if RANK != -1: |
|
train_loader.sampler.set_epoch(epoch) |
|
pbar = enumerate(train_loader) |
|
LOGGER.info( |
|
("\n" + "%10s" * 7) |
|
% ("Epoch", "gpu_mem", "box", "obj", "cls", "labels", "img_size") |
|
) |
|
if RANK in [-1, 0]: |
|
pbar = tqdm(pbar, total=nb) |
|
optimizer.zero_grad() |
|
for i, ( |
|
imgs, |
|
targets, |
|
paths, |
|
_, |
|
) in ( |
|
pbar |
|
): |
|
ni = ( |
|
i + nb * epoch |
|
) |
|
imgs = ( |
|
imgs.to(device, non_blocking=True).float() / 255.0 |
|
) |
|
|
|
|
|
if ni <= nw: |
|
xi = [0, nw] |
|
|
|
accumulate = max( |
|
1, np.interp(ni, xi, [1, nbs / batch_size]).round() |
|
) |
|
for j, x in enumerate(optimizer.param_groups): |
|
|
|
x["lr"] = np.interp( |
|
ni, |
|
xi, |
|
[ |
|
hyp["warmup_bias_lr"] if j == 2 else 0.0, |
|
x["initial_lr"] * lf(epoch), |
|
], |
|
) |
|
if "momentum" in x: |
|
x["momentum"] = np.interp( |
|
ni, xi, [hyp["warmup_momentum"], hyp["momentum"]] |
|
) |
|
|
|
|
|
if opt.multi_scale: |
|
sz = ( |
|
random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs |
|
) |
|
sf = sz / max(imgs.shape[2:]) |
|
if sf != 1: |
|
ns = [ |
|
math.ceil(x * sf / gs) * gs for x in imgs.shape[2:] |
|
] |
|
imgs = nn.functional.interpolate( |
|
imgs, size=ns, mode="bilinear", align_corners=False |
|
) |
|
|
|
|
|
with amp.autocast(enabled=cuda): |
|
pred = model(imgs) |
|
loss, loss_items = compute_loss( |
|
pred, targets.to(device) |
|
) |
|
if RANK != -1: |
|
loss *= WORLD_SIZE |
|
if opt.quad: |
|
loss *= 4.0 |
|
|
|
|
|
scaler.scale(loss).backward() |
|
|
|
|
|
if ni - last_opt_step >= accumulate: |
|
scaler.step(optimizer) |
|
scaler.update() |
|
optimizer.zero_grad() |
|
if ema: |
|
ema.update(model) |
|
last_opt_step = ni |
|
|
|
|
|
if RANK in [-1, 0]: |
|
mloss = (mloss * i + loss_items) / ( |
|
i + 1 |
|
) |
|
mem = f"{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G" |
|
pbar.set_description( |
|
("%10s" * 2 + "%10.4g" * 5) |
|
% ( |
|
f"{epoch}/{epochs - 1}", |
|
mem, |
|
*mloss, |
|
targets.shape[0], |
|
imgs.shape[-1], |
|
) |
|
) |
|
callbacks.run( |
|
"on_train_batch_end", |
|
ni, |
|
model, |
|
imgs, |
|
targets, |
|
paths, |
|
plots, |
|
opt.sync_bn, |
|
) |
|
|
|
|
|
|
|
lr = [x["lr"] for x in optimizer.param_groups] |
|
scheduler.step() |
|
|
|
if RANK in [-1, 0]: |
|
|
|
callbacks.run("on_train_epoch_end", epoch=epoch) |
|
ema.update_attr( |
|
model, |
|
include=[ |
|
"yaml", |
|
"nc", |
|
"hyp", |
|
"names", |
|
"stride", |
|
"class_weights", |
|
], |
|
) |
|
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop |
|
if not noval or final_epoch: |
|
results, maps, _ = val.run( |
|
data_dict, |
|
batch_size=batch_size // WORLD_SIZE * 2, |
|
imgsz=imgsz, |
|
model=ema.ema, |
|
single_cls=single_cls, |
|
dataloader=val_loader, |
|
save_dir=save_dir, |
|
save_json=is_coco and final_epoch, |
|
verbose=nc < 50 and final_epoch, |
|
plots=plots and final_epoch, |
|
callbacks=callbacks, |
|
compute_loss=compute_loss, |
|
) |
|
|
|
|
|
fi = fitness( |
|
np.array(results).reshape(1, -1) |
|
) |
|
if fi > best_fitness: |
|
best_fitness = fi |
|
log_vals = list(mloss) + list(results) + lr |
|
callbacks.run( |
|
"on_fit_epoch_end", log_vals, epoch, best_fitness, fi |
|
) |
|
|
|
|
|
if (not nosave) or (final_epoch and not evolve): |
|
ckpt = { |
|
"epoch": epoch, |
|
"best_fitness": best_fitness, |
|
"model": deepcopy(de_parallel(model)).half(), |
|
"ema": deepcopy(ema.ema).half(), |
|
"updates": ema.updates, |
|
"optimizer": optimizer.state_dict(), |
|
"wandb_id": loggers.wandb.wandb_run.id |
|
if loggers.wandb |
|
else None, |
|
} |
|
|
|
|
|
torch.save(ckpt, last) |
|
if best_fitness == fi: |
|
torch.save(ckpt, best) |
|
del ckpt |
|
callbacks.run( |
|
"on_model_save", last, epoch, final_epoch, best_fitness, fi |
|
) |
|
|
|
|
|
if RANK == -1 and stopper(epoch=epoch, fitness=fi): |
|
break |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if RANK in [-1, 0]: |
|
LOGGER.info( |
|
f"\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours." |
|
) |
|
if not evolve: |
|
if is_coco: |
|
for m in ( |
|
[last, best] if best.exists() else [last] |
|
): |
|
results, _, _ = val.run( |
|
data_dict, |
|
batch_size=batch_size // WORLD_SIZE * 2, |
|
imgsz=imgsz, |
|
model=attempt_load(m, device).half(), |
|
iou_thres=0.7, |
|
single_cls=single_cls, |
|
dataloader=val_loader, |
|
save_dir=save_dir, |
|
save_json=True, |
|
plots=False, |
|
) |
|
|
|
for f in last, best: |
|
if f.exists(): |
|
strip_optimizer(f) |
|
callbacks.run("on_train_end", last, best, plots, epoch) |
|
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}") |
|
|
|
torch.cuda.empty_cache() |
|
return results |
|
|
|
|
|
def parse_opt(known=False): |
|
parser = argparse.ArgumentParser() |
|
parser.add_argument( |
|
"--weights", |
|
type=str, |
|
default="yolov5s.pt", |
|
help="initial weights path", |
|
) |
|
parser.add_argument("--cfg", type=str, default="", help="model.yaml path") |
|
parser.add_argument( |
|
"--data", |
|
type=str, |
|
default="data/coco128.yaml", |
|
help="dataset.yaml path", |
|
) |
|
parser.add_argument( |
|
"--hyp", |
|
type=str, |
|
default="data/hyps/hyp.scratch.yaml", |
|
help="hyperparameters path", |
|
) |
|
parser.add_argument("--epochs", type=int, default=300) |
|
parser.add_argument( |
|
"--batch-size", |
|
type=int, |
|
default=16, |
|
help="total batch size for all GPUs", |
|
) |
|
parser.add_argument( |
|
"--imgsz", |
|
"--img", |
|
"--img-size", |
|
type=int, |
|
default=640, |
|
help="train, val image size (pixels)", |
|
) |
|
parser.add_argument( |
|
"--rect", action="store_true", help="rectangular training" |
|
) |
|
parser.add_argument( |
|
"--resume", |
|
nargs="?", |
|
const=True, |
|
default=False, |
|
help="resume most recent training", |
|
) |
|
parser.add_argument( |
|
"--nosave", action="store_true", help="only save final checkpoint" |
|
) |
|
parser.add_argument( |
|
"--noval", action="store_true", help="only validate final epoch" |
|
) |
|
parser.add_argument( |
|
"--noautoanchor", action="store_true", help="disable autoanchor check" |
|
) |
|
parser.add_argument( |
|
"--evolve", |
|
type=int, |
|
nargs="?", |
|
const=300, |
|
help="evolve hyperparameters for x generations", |
|
) |
|
parser.add_argument("--bucket", type=str, default="", help="gsutil bucket") |
|
parser.add_argument( |
|
"--cache", |
|
type=str, |
|
nargs="?", |
|
const="ram", |
|
help='--cache images in "ram" (default) or "disk"', |
|
) |
|
parser.add_argument( |
|
"--image-weights", |
|
action="store_true", |
|
help="use weighted image selection for training", |
|
) |
|
parser.add_argument( |
|
"--device", default="", help="cuda device, i.e. 0 or 0,1,2,3 or cpu" |
|
) |
|
parser.add_argument( |
|
"--multi-scale", action="store_true", help="vary img-size +/- 50%%" |
|
) |
|
parser.add_argument( |
|
"--single-cls", |
|
action="store_true", |
|
help="train multi-class data as single-class", |
|
) |
|
parser.add_argument( |
|
"--adam", action="store_true", help="use torch.optim.Adam() optimizer" |
|
) |
|
parser.add_argument( |
|
"--sync-bn", |
|
action="store_true", |
|
help="use SyncBatchNorm, only available in DDP mode", |
|
) |
|
parser.add_argument( |
|
"--workers", |
|
type=int, |
|
default=8, |
|
help="maximum number of dataloader workers", |
|
) |
|
parser.add_argument( |
|
"--project", default="runs/train", help="save to project/name" |
|
) |
|
parser.add_argument("--entity", default=None, help="W&B entity") |
|
parser.add_argument("--name", default="exp", help="save to project/name") |
|
parser.add_argument( |
|
"--exist-ok", |
|
action="store_true", |
|
help="existing project/name ok, do not increment", |
|
) |
|
parser.add_argument("--quad", action="store_true", help="quad dataloader") |
|
parser.add_argument("--linear-lr", action="store_true", help="linear LR") |
|
parser.add_argument( |
|
"--label-smoothing", |
|
type=float, |
|
default=0.0, |
|
help="Label smoothing epsilon", |
|
) |
|
parser.add_argument( |
|
"--upload_dataset", |
|
action="store_true", |
|
help="Upload dataset as W&B artifact table", |
|
) |
|
parser.add_argument( |
|
"--bbox_interval", |
|
type=int, |
|
default=-1, |
|
help="Set bounding-box image logging interval for W&B", |
|
) |
|
parser.add_argument( |
|
"--save_period", |
|
type=int, |
|
default=-1, |
|
help='Log model after every "save_period" epoch', |
|
) |
|
parser.add_argument( |
|
"--artifact_alias", |
|
type=str, |
|
default="latest", |
|
help="version of dataset artifact to be used", |
|
) |
|
parser.add_argument( |
|
"--local_rank", |
|
type=int, |
|
default=-1, |
|
help="DDP parameter, do not modify", |
|
) |
|
parser.add_argument( |
|
"--freeze", |
|
type=int, |
|
default=0, |
|
help="Number of layers to freeze. backbone=10, all=24", |
|
) |
|
parser.add_argument( |
|
"--patience", |
|
type=int, |
|
default=100, |
|
help="EarlyStopping patience (epochs without improvement)", |
|
) |
|
opt = parser.parse_known_args()[0] if known else parser.parse_args() |
|
return opt |
|
|
|
|
|
def main(opt, callbacks=Callbacks()): |
|
|
|
set_logging(RANK) |
|
if RANK in [-1, 0]: |
|
print( |
|
colorstr("train: ") |
|
+ ", ".join(f"{k}={v}" for k, v in vars(opt).items()) |
|
) |
|
check_git_status() |
|
check_requirements( |
|
requirements=FILE.parent / "requirements.txt", exclude=["thop"] |
|
) |
|
|
|
|
|
if ( |
|
opt.resume and not check_wandb_resume(opt) and not opt.evolve |
|
): |
|
ckpt = ( |
|
opt.resume if isinstance(opt.resume, str) else get_latest_run() |
|
) |
|
assert os.path.isfile( |
|
ckpt |
|
), "ERROR: --resume checkpoint does not exist" |
|
with open(Path(ckpt).parent.parent / "opt.yaml") as f: |
|
opt = argparse.Namespace(**yaml.safe_load(f)) |
|
opt.cfg, opt.weights, opt.resume = "", ckpt, True |
|
LOGGER.info(f"Resuming training from {ckpt}") |
|
else: |
|
opt.data, opt.cfg, opt.hyp = ( |
|
check_file(opt.data), |
|
check_yaml(opt.cfg), |
|
check_yaml(opt.hyp), |
|
) |
|
assert len(opt.cfg) or len( |
|
opt.weights |
|
), "either --cfg or --weights must be specified" |
|
if opt.evolve: |
|
opt.project = "runs/evolve" |
|
opt.exist_ok = opt.resume |
|
opt.save_dir = str( |
|
increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok) |
|
) |
|
|
|
|
|
device = select_device(opt.device, batch_size=opt.batch_size) |
|
if LOCAL_RANK != -1: |
|
from datetime import timedelta |
|
|
|
assert ( |
|
torch.cuda.device_count() > LOCAL_RANK |
|
), "insufficient CUDA devices for DDP command" |
|
assert ( |
|
opt.batch_size % WORLD_SIZE == 0 |
|
), "--batch-size must be multiple of CUDA device count" |
|
assert ( |
|
not opt.image_weights |
|
), "--image-weights argument is not compatible with DDP training" |
|
assert ( |
|
not opt.evolve |
|
), "--evolve argument is not compatible with DDP training" |
|
torch.cuda.set_device(LOCAL_RANK) |
|
device = torch.device("cuda", LOCAL_RANK) |
|
dist.init_process_group( |
|
backend="nccl" if dist.is_nccl_available() else "gloo" |
|
) |
|
|
|
|
|
if not opt.evolve: |
|
train(opt.hyp, opt, device, callbacks) |
|
if WORLD_SIZE > 1 and RANK == 0: |
|
_ = [ |
|
print("Destroying process group... ", end=""), |
|
dist.destroy_process_group(), |
|
print("Done."), |
|
] |
|
|
|
|
|
else: |
|
|
|
meta = { |
|
"lr0": ( |
|
1, |
|
1e-5, |
|
1e-1, |
|
), |
|
"lrf": ( |
|
1, |
|
0.01, |
|
1.0, |
|
), |
|
"momentum": (0.3, 0.6, 0.98), |
|
"weight_decay": (1, 0.0, 0.001), |
|
"warmup_epochs": (1, 0.0, 5.0), |
|
"warmup_momentum": (1, 0.0, 0.95), |
|
"warmup_bias_lr": (1, 0.0, 0.2), |
|
"box": (1, 0.02, 0.2), |
|
"cls": (1, 0.2, 4.0), |
|
"cls_pw": (1, 0.5, 2.0), |
|
"obj": (1, 0.2, 4.0), |
|
"obj_pw": (1, 0.5, 2.0), |
|
"iou_t": (0, 0.1, 0.7), |
|
"anchor_t": (1, 2.0, 8.0), |
|
"anchors": (2, 2.0, 10.0), |
|
"fl_gamma": ( |
|
0, |
|
0.0, |
|
2.0, |
|
), |
|
"hsv_h": (1, 0.0, 0.1), |
|
"hsv_s": ( |
|
1, |
|
0.0, |
|
0.9, |
|
), |
|
"hsv_v": (1, 0.0, 0.9), |
|
"degrees": (1, 0.0, 45.0), |
|
"translate": (1, 0.0, 0.9), |
|
"scale": (1, 0.0, 0.9), |
|
"shear": (1, 0.0, 10.0), |
|
"perspective": ( |
|
0, |
|
0.0, |
|
0.001, |
|
), |
|
"flipud": (1, 0.0, 1.0), |
|
"fliplr": (0, 0.0, 1.0), |
|
"mosaic": (1, 0.0, 1.0), |
|
"mixup": (1, 0.0, 1.0), |
|
"copy_paste": (1, 0.0, 1.0), |
|
} |
|
|
|
with open(opt.hyp) as f: |
|
hyp = yaml.safe_load(f) |
|
if "anchors" not in hyp: |
|
hyp["anchors"] = 3 |
|
opt.noval, opt.nosave, save_dir = ( |
|
True, |
|
True, |
|
Path(opt.save_dir), |
|
) |
|
|
|
evolve_yaml, evolve_csv = ( |
|
save_dir / "hyp_evolve.yaml", |
|
save_dir / "evolve.csv", |
|
) |
|
if opt.bucket: |
|
os.system( |
|
f"gsutil cp gs://{opt.bucket}/evolve.csv {save_dir}" |
|
) |
|
|
|
for _ in range(opt.evolve): |
|
if ( |
|
evolve_csv.exists() |
|
): |
|
|
|
parent = ( |
|
"single" |
|
) |
|
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=",", skiprows=1) |
|
n = min(5, len(x)) |
|
x = x[np.argsort(-fitness(x))][:n] |
|
w = fitness(x) - fitness(x).min() + 1e-6 |
|
if parent == "single" or len(x) == 1: |
|
|
|
x = x[ |
|
random.choices(range(n), weights=w)[0] |
|
] |
|
elif parent == "weighted": |
|
x = (x * w.reshape(n, 1)).sum( |
|
0 |
|
) / w.sum() |
|
|
|
|
|
mp, s = 0.8, 0.2 |
|
npr = np.random |
|
npr.seed(int(time.time())) |
|
g = np.array([meta[k][0] for k in hyp.keys()]) |
|
ng = len(meta) |
|
v = np.ones(ng) |
|
while all( |
|
v == 1 |
|
): |
|
v = ( |
|
g |
|
* (npr.random(ng) < mp) |
|
* npr.randn(ng) |
|
* npr.random() |
|
* s |
|
+ 1 |
|
).clip(0.3, 3.0) |
|
for i, k in enumerate(hyp.keys()): |
|
hyp[k] = float(x[i + 7] * v[i]) |
|
|
|
|
|
for k, v in meta.items(): |
|
hyp[k] = max(hyp[k], v[1]) |
|
hyp[k] = min(hyp[k], v[2]) |
|
hyp[k] = round(hyp[k], 5) |
|
|
|
|
|
results = train(hyp.copy(), opt, device, callbacks) |
|
|
|
|
|
print_mutation(results, hyp.copy(), save_dir, opt.bucket) |
|
|
|
|
|
plot_evolve(evolve_csv) |
|
print( |
|
f"Hyperparameter evolution finished\n" |
|
f"Results saved to {colorstr('bold', save_dir)}\n" |
|
f"Use best hyperparameters example: $ python train.py --hyp {evolve_yaml}" |
|
) |
|
|
|
|
|
def run(**kwargs): |
|
|
|
opt = parse_opt(True) |
|
for k, v in kwargs.items(): |
|
setattr(opt, k, v) |
|
main(opt) |
|
|
|
|
|
if __name__ == "__main__": |
|
opt = parse_opt() |
|
main(opt) |
|
|