ultralytics 8.0.167 Tuner updates and HUB Pose and Classify fixes (#4656)
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
Glenn Jocher
GitHub
parent
8596ee241f
commit
d2cf7acce0
21 changed files with 174 additions and 144 deletions
|
|
@ -438,18 +438,16 @@ class Exporter:
|
|||
Path(asset).unlink() # delete zip
|
||||
pnnx.chmod(0o777) # set read, write, and execute permissions for everyone
|
||||
|
||||
use_ncnn = True
|
||||
ncnn_args = [
|
||||
f'ncnnparam={f / "model.ncnn.param"}',
|
||||
f'ncnnbin={f / "model.ncnn.bin"}',
|
||||
f'ncnnpy={f / "model_ncnn.py"}', ] if use_ncnn else []
|
||||
f'ncnnpy={f / "model_ncnn.py"}', ]
|
||||
|
||||
use_pnnx = False
|
||||
pnnx_args = [
|
||||
f'pnnxparam={f / "model.pnnx.param"}',
|
||||
f'pnnxbin={f / "model.pnnx.bin"}',
|
||||
f'pnnxpy={f / "model_pnnx.py"}',
|
||||
f'pnnxonnx={f / "model.pnnx.onnx"}', ] if use_pnnx else []
|
||||
f'pnnxonnx={f / "model.pnnx.onnx"}', ]
|
||||
|
||||
cmd = [
|
||||
str(pnnx),
|
||||
|
|
@ -462,7 +460,10 @@ class Exporter:
|
|||
f.mkdir(exist_ok=True) # make ncnn_model directory
|
||||
LOGGER.info(f"{prefix} running '{' '.join(cmd)}'")
|
||||
subprocess.run(cmd, check=True)
|
||||
for f_debug in 'debug.bin', 'debug.param', 'debug2.bin', 'debug2.param': # remove debug files
|
||||
|
||||
# Remove debug files
|
||||
pnnx_files = [x.split('=')[-1] for x in pnnx_args]
|
||||
for f_debug in ('debug.bin', 'debug.param', 'debug2.bin', 'debug2.param', *pnnx_files):
|
||||
Path(f_debug).unlink(missing_ok=True)
|
||||
|
||||
yaml_save(f / 'metadata.yaml', self.metadata) # add metadata.yaml
|
||||
|
|
|
|||
|
|
@ -341,7 +341,8 @@ class Model:
|
|||
self.trainer.train()
|
||||
# Update model and cfg after training
|
||||
if RANK in (-1, 0):
|
||||
self.model, _ = attempt_load_one_weight(str(self.trainer.best))
|
||||
ckpt = self.trainer.best if self.trainer.best.exists() else self.trainer.last
|
||||
self.model, _ = attempt_load_one_weight(ckpt)
|
||||
self.overrides = self.model.args
|
||||
self.metrics = getattr(self.trainer.validator, 'metrics', None) # TODO: no metrics returned by DDP
|
||||
return self.metrics
|
||||
|
|
@ -360,9 +361,9 @@ class Model:
|
|||
else:
|
||||
from .tuner import Tuner
|
||||
|
||||
custom = {} # method defaults
|
||||
custom = {'plots': False, 'save': False} # method defaults
|
||||
args = {**self.overrides, **custom, **kwargs, 'mode': 'train'} # highest priority args on the right
|
||||
return Tuner(args=args, _callbacks=self.callbacks)(model=self.model, iterations=iterations)
|
||||
return Tuner(args=args, _callbacks=self.callbacks)(model=self, iterations=iterations)
|
||||
|
||||
def to(self, device):
|
||||
"""
|
||||
|
|
|
|||
|
|
@ -115,7 +115,7 @@ class BaseTrainer:
|
|||
try:
|
||||
if self.args.task == 'classify':
|
||||
self.data = check_cls_dataset(self.args.data)
|
||||
elif self.args.data.split('.')[-1] in ('yaml', 'yml') or self.args.task in ('detect', 'segment'):
|
||||
elif self.args.data.split('.')[-1] in ('yaml', 'yml') or self.args.task in ('detect', 'segment', 'pose'):
|
||||
self.data = check_det_dataset(self.args.data)
|
||||
if 'yaml_file' in self.data:
|
||||
self.args.data = self.data['yaml_file'] # for validating 'yolo train data=url.zip' usage
|
||||
|
|
@ -251,9 +251,8 @@ class BaseTrainer:
|
|||
self.args.imgsz = check_imgsz(self.args.imgsz, stride=gs, floor=gs, max_dim=1)
|
||||
|
||||
# Batch size
|
||||
if self.batch_size == -1:
|
||||
if RANK == -1: # single-GPU only, estimate best batch size
|
||||
self.args.batch = self.batch_size = check_train_batch_size(self.model, self.args.imgsz, self.amp)
|
||||
if self.batch_size == -1 and RANK == -1: # single-GPU only, estimate best batch size
|
||||
self.args.batch = self.batch_size = check_train_batch_size(self.model, self.args.imgsz, self.amp)
|
||||
|
||||
# Dataloaders
|
||||
batch_size = self.batch_size // max(world_size, 1)
|
||||
|
|
@ -262,7 +261,7 @@ class BaseTrainer:
|
|||
self.test_loader = self.get_dataloader(self.testset, batch_size=batch_size * 2, rank=-1, mode='val')
|
||||
self.validator = self.get_validator()
|
||||
metric_keys = self.validator.metrics.keys + self.label_loss_items(prefix='val')
|
||||
self.metrics = dict(zip(metric_keys, [0] * len(metric_keys))) # TODO: init metrics for plot_results()?
|
||||
self.metrics = dict(zip(metric_keys, [0] * len(metric_keys)))
|
||||
self.ema = ModelEMA(self.model)
|
||||
if self.args.plots:
|
||||
self.plot_training_labels()
|
||||
|
|
|
|||
|
|
@ -18,6 +18,7 @@ Example:
|
|||
"""
|
||||
import random
|
||||
import time
|
||||
from copy import deepcopy
|
||||
|
||||
import numpy as np
|
||||
|
||||
|
|
@ -51,7 +52,7 @@ class Tuner:
|
|||
from ultralytics import YOLO
|
||||
|
||||
model = YOLO('yolov8n.pt')
|
||||
model.tune(data='coco8.yaml', imgsz=640, epochs=100, iterations=10)
|
||||
model.tune(data='coco8.yaml', imgsz=640, epochs=100, iterations=10, val=False, cache=True)
|
||||
```
|
||||
"""
|
||||
|
||||
|
|
@ -63,11 +64,11 @@ class Tuner:
|
|||
args (dict, optional): Configuration for hyperparameter evolution.
|
||||
"""
|
||||
self.args = get_cfg(overrides=args)
|
||||
self.space = {
|
||||
self.space = { # key: (min, max, gain(optionaL))
|
||||
# 'optimizer': tune.choice(['SGD', 'Adam', 'AdamW', 'NAdam', 'RAdam', 'RMSProp']),
|
||||
'lr0': (1e-5, 1e-1),
|
||||
'lrf': (0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
|
||||
'momentum': (0.6, 0.98), # SGD momentum/Adam beta1
|
||||
'momentum': (0.6, 0.98, 0.3), # SGD momentum/Adam beta1
|
||||
'weight_decay': (0.0, 0.001), # optimizer weight decay 5e-4
|
||||
'warmup_epochs': (0.0, 5.0), # warmup epochs (fractions ok)
|
||||
'warmup_momentum': (0.0, 0.95), # warmup initial momentum
|
||||
|
|
@ -86,13 +87,13 @@ class Tuner:
|
|||
'mosaic': (0.0, 1.0), # image mixup (probability)
|
||||
'mixup': (0.0, 1.0), # image mixup (probability)
|
||||
'copy_paste': (0.0, 1.0)} # segment copy-paste (probability)
|
||||
self.tune_dir = get_save_dir(self.args, name='tune')
|
||||
self.tune_dir = get_save_dir(self.args, name='_tune')
|
||||
self.evolve_csv = self.tune_dir / 'evolve.csv'
|
||||
self.callbacks = _callbacks or callbacks.get_default_callbacks()
|
||||
callbacks.add_integration_callbacks(self)
|
||||
LOGGER.info(f"Initialized Tuner instance with 'tune_dir={self.tune_dir}'.")
|
||||
|
||||
def _mutate(self, parent='single', n=5, mutation=0.8, sigma=0.2, return_best=False):
|
||||
def _mutate(self, parent='single', n=5, mutation=0.8, sigma=0.2):
|
||||
"""
|
||||
Mutates the hyperparameters based on bounds and scaling factors specified in `self.space`.
|
||||
|
||||
|
|
@ -111,10 +112,7 @@ class Tuner:
|
|||
fitness = x[:, 0] # first column
|
||||
n = min(n, len(x)) # number of previous results to consider
|
||||
x = x[np.argsort(-fitness)][:n] # top n mutations
|
||||
if return_best:
|
||||
return {k: float(x[0, i + 1]) for i, k in enumerate(self.space.keys())}
|
||||
fitness = x[:, 0] # first column
|
||||
w = fitness - fitness.min() + 1E-6 # weights (sum > 0)
|
||||
w = x[:, 0] - x[:, 0].min() + 1E-6 # weights (sum > 0)
|
||||
if parent == 'single' or len(x) == 1:
|
||||
# x = x[random.randint(0, n - 1)] # random selection
|
||||
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
|
||||
|
|
@ -124,7 +122,7 @@ class Tuner:
|
|||
# Mutate
|
||||
r = np.random # method
|
||||
r.seed(int(time.time()))
|
||||
g = np.array([self.space[k][0] for k in self.space.keys()]) # gains 0-1
|
||||
g = np.array([v[2] if len(v) == 3 else 1.0 for k, v in self.space.items()]) # gains 0-1
|
||||
ng = len(self.space)
|
||||
v = np.ones(ng)
|
||||
while all(v == 1): # mutate until a change occurs (prevent duplicates)
|
||||
|
|
@ -152,7 +150,7 @@ class Tuner:
|
|||
4. Log the fitness score and mutated hyperparameters to a CSV file.
|
||||
|
||||
Args:
|
||||
model (YOLO): A pre-initialized YOLO model to be used for training.
|
||||
model (Model): A pre-initialized YOLO model to be used for training.
|
||||
iterations (int): The number of generations to run the evolution for.
|
||||
|
||||
Note:
|
||||
|
|
@ -160,6 +158,7 @@ class Tuner:
|
|||
Ensure this path is set correctly in the Tuner instance.
|
||||
"""
|
||||
|
||||
t0 = time.time()
|
||||
self.tune_dir.mkdir(parents=True, exist_ok=True)
|
||||
for i in range(iterations):
|
||||
# Mutate hyperparameters
|
||||
|
|
@ -167,17 +166,27 @@ class Tuner:
|
|||
LOGGER.info(f'{prefix} Starting iteration {i + 1}/{iterations} with hyperparameters: {mutated_hyp}')
|
||||
|
||||
# Initialize and train YOLOv8 model
|
||||
model = YOLO('yolov8n.pt')
|
||||
train_args = {**vars(self.args), **mutated_hyp}
|
||||
results = model.train(**train_args)
|
||||
try:
|
||||
train_args = {**vars(self.args), **mutated_hyp}
|
||||
fitness = (deepcopy(model) or YOLO(self.args.model)).train(**train_args).fitness # results.fitness
|
||||
except Exception as e:
|
||||
LOGGER.warning(f'WARNING ❌️ training failure for hyperparameter tuning iteration {i}\n{e}')
|
||||
fitness = 0.0
|
||||
|
||||
# Save results and mutated_hyp to evolve_csv
|
||||
log_row = [round(fitness, 5)] + [mutated_hyp[k] for k in self.space.keys()]
|
||||
headers = '' if self.evolve_csv.exists() else (','.join(['fitness_score'] + list(self.space.keys())) + '\n')
|
||||
log_row = [results.fitness] + [mutated_hyp[k] for k in self.space.keys()]
|
||||
with open(self.evolve_csv, 'a') as f:
|
||||
f.write(headers + ','.join(map(str, log_row)) + '\n')
|
||||
|
||||
LOGGER.info(f'{prefix} All iterations complete. Results saved to {colorstr("bold", self.tune_dir)}')
|
||||
best_hyp = self._mutate(return_best=True) # best hyps
|
||||
yaml_save(self.tune_dir / 'best.yaml', best_hyp)
|
||||
# Print tuning results
|
||||
x = np.loadtxt(self.evolve_csv, ndmin=2, delimiter=',', skiprows=1)
|
||||
fitness = x[:, 0] # first column
|
||||
i = np.argsort(-fitness)[0] # best fitness index
|
||||
LOGGER.info(f'\n{prefix} All iterations complete ✅ ({time.time() - t0:.2f}s)\n'
|
||||
f'{prefix} Results saved to {colorstr("bold", self.tune_dir)}\n'
|
||||
f'{prefix} Best fitness={fitness[i]} observed at iteration {i}')
|
||||
|
||||
# Save turning results
|
||||
yaml_save(self.tune_dir / 'best.yaml', data={k: float(x[0, i + 1]) for i, k in enumerate(self.space.keys())})
|
||||
yaml_print(self.tune_dir / 'best.yaml')
|
||||
|
|
|
|||
|
|
@ -111,12 +111,12 @@ class BaseValidator:
|
|||
if self.training:
|
||||
self.device = trainer.device
|
||||
self.data = trainer.data
|
||||
model = trainer.ema.ema or trainer.model
|
||||
self.args.half = self.device.type != 'cpu' # force FP16 val during training
|
||||
model = trainer.ema.ema or trainer.model
|
||||
model = model.half() if self.args.half else model.float()
|
||||
self.model = model
|
||||
# self.model = model
|
||||
self.loss = torch.zeros_like(trainer.loss_items, device=trainer.device)
|
||||
self.args.plots = trainer.stopper.possible_stop or (trainer.epoch == trainer.epochs - 1)
|
||||
self.args.plots &= trainer.stopper.possible_stop or (trainer.epoch == trainer.epochs - 1)
|
||||
model.eval()
|
||||
else:
|
||||
callbacks.add_integration_callbacks(self)
|
||||
|
|
@ -126,7 +126,7 @@ class BaseValidator:
|
|||
dnn=self.args.dnn,
|
||||
data=self.args.data,
|
||||
fp16=self.args.half)
|
||||
self.model = model
|
||||
# self.model = model
|
||||
self.device = model.device # update device
|
||||
self.args.half = model.fp16 # update half
|
||||
stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
|
||||
|
|
@ -297,8 +297,7 @@ class BaseValidator:
|
|||
|
||||
def on_plot(self, name, data=None):
|
||||
"""Registers plots (e.g. to be consumed in callbacks)"""
|
||||
path = Path(name)
|
||||
self.plots[path] = {'data': data, 'timestamp': time.time()}
|
||||
self.plots[Path(name)] = {'data': data, 'timestamp': time.time()}
|
||||
|
||||
# TODO: may need to put these following functions into callback
|
||||
def plot_val_samples(self, batch, ni):
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue