File size: 8,239 Bytes
bc06f47 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 |
import config
import torch
import torch.optim as optim
from model import YOLOv3
from tqdm import tqdm
from utils import (
mean_average_precision,
cells_to_bboxes,
get_evaluation_bboxes,
save_checkpoint,
load_checkpoint,
check_class_accuracy,
get_loaders,
plot_couple_examples
)
from torch.utils.data import DataLoader
from loss import YoloLoss
import warnings
warnings.filterwarnings("ignore")
from pytorch_lightning import LightningModule
from pytorch_lightning.callbacks import LearningRateMonitor
from pytorch_lightning.callbacks.progress import TQDMProgressBar
from torch.optim.lr_scheduler import OneCycleLR
from torchmetrics.functional import accuracy
class LitYolo(LightningModule):
def __init__(self, batch_size=64):
super().__init__()
self.lr = config.LEARNING_RATE
self.weight_decay =config.WEIGHT_DECAY
self.model = YOLOv3(num_classes=config.NUM_CLASSES)
self.save_hyperparameters()
self.optimizer = optim.Adam(
self.model.parameters(), lr=self.lr, weight_decay=self.weight_decay
)
self.scaler = torch.cuda.amp.GradScaler()
self.loss_fn = YoloLoss()
self.scaled_anchors = (
torch.tensor(config.ANCHORS)
* torch.tensor(config.S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
).to(config.DEVICE)
self.losses =[]
def forward(self, x):
return self.model(x)
def training_step(self, batch, batch_idx):
x, y = batch
x = x.to(config.DEVICE)
y0, y1, y2 = (
y[0].to(config.DEVICE),
y[1].to(config.DEVICE),
y[2].to(config.DEVICE),
)
with torch.cuda.amp.autocast():
out = self.model(x)
loss = (
self.loss_fn(out[0], y0, self.scaled_anchors[0])
+ self.loss_fn(out[1], y1, self.scaled_anchors[1])
+ self.loss_fn(out[2], y2, self.scaled_anchors[2])
)
self.losses.append(loss.item())
self.optimizer.zero_grad()
self.scaler.scale(loss).backward(retain_graph=True)
self.scaler.step(self.optimizer)
self.scaler.update()
mean_loss = sum(self.losses) / len(self.losses)
# Calling self.log will surface up scalars for you in TensorBoard
self.log("mean_loss = ", mean_loss, prog_bar=True)
return loss
#def validation_step(self, batch, batch_idx):
# pass
#def test_step(self, batch, batch_idx):
# pass
def on_train_epoch_end(self):
epoch = self.trainer.current_epoch + 1
print(f"Currently epoch {epoch-1}")
if config.SAVE_MODEL:
save_checkpoint(self.model, self.optimizer, filename=config.CHECKPOINT_FILE)
if epoch > 1 and epoch % 10 == 0
plot_couple_examples(self.model, self.test_dataloader(), 0.6, 0.5, self.scaled_anchors)
print(f"Currently epoch {epoch-1}")
print("On Train loader:")
check_class_accuracy(self.model, self.train_dataloader(), threshold=config.CONF_THRESHOLD)
if epoch > 30 and epoch % 8 == 0:
check_class_accuracy(self.model, self.test_dataloader(), threshold=config.CONF_THRESHOLD)
pred_boxes, true_boxes = get_evaluation_bboxes(
self.test_dataloader(),
self.model,
iou_threshold=config.NMS_IOU_THRESH,
anchors=config.ANCHORS,
threshold=config.CONF_THRESHOLD,
)
mapval = mean_average_precision(
pred_boxes,
true_boxes,
iou_threshold=config.MAP_IOU_THRESH,
box_format="midpoint",
num_classes=config.NUM_CLASSES,
)
print(f"MAP: {mapval.item()}")
self.losses =[]
self.model.train()
def lr_finder(self, num_iter=50):
from torch_lr_finder import LRFinder
def criterion(out, y):
y0, y1, y2 = (
y[0].to(config.DEVICE),
y[1].to(config.DEVICE),
y[2].to(config.DEVICE),
)
loss = (
self.loss_fn(out[0], y0, self.scaled_anchors[0])
+ self.loss_fn(out[1], y1, self.scaled_anchors[1])
+ self.loss_fn(out[2], y2, self.scaled_anchors[2])
)
return loss
lr_finder = LRFinder(self.model, self.optimizer, criterion, device=config.DEVICE)
lr_finder.range_test(self.train_dataloader(), end_lr=1, num_iter=num_iter, step_mode="exp")
ax, suggested_lr = lr_finder.plot() # to inspect the loss-learning rate graph
lr_finder.reset() # to reset the model and optimizer to their initial state
return suggested_lr
def configure_optimizers(self):
#suggested_lr = self.lr_finder() #check on self.train_dataloader
suggested_lr = 6.25E-03
steps_per_epoch = len(self.train_dataloader())
scheduler_dict = {
"scheduler": OneCycleLR(
self.optimizer, max_lr=suggested_lr,
steps_per_epoch=steps_per_epoch,
epochs=self.trainer.max_epochs,
pct_start=5/self.trainer.max_epochs,
three_phase=False,
div_factor=80,
final_div_factor=400,
anneal_strategy='linear',
),
"interval": "step",
}
return {"optimizer": self.optimizer,"lr_scheduler": scheduler_dict} #
####################
# DATA RELATED HOOKS
####################
def prepare_data(self):
# download
from dataset import YOLODataset
IMAGE_SIZE = config.IMAGE_SIZE
train_csv_path=config.DATASET + "/train.csv"
test_csv_path=config.DATASET + "/test.csv"
self.train_dataset = YOLODataset(
train_csv_path,
transform=config.train_transforms,
S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
img_dir=config.IMG_DIR,
label_dir=config.LABEL_DIR,
anchors=config.ANCHORS,
)
self.test_dataset = YOLODataset(
test_csv_path,
transform=config.test_transforms,
S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
img_dir=config.IMG_DIR,
label_dir=config.LABEL_DIR,
anchors=config.ANCHORS,
)
self.val_dataset = YOLODataset(
train_csv_path,
transform=config.test_transforms,
S=[IMAGE_SIZE // 32, IMAGE_SIZE // 16, IMAGE_SIZE // 8],
img_dir=config.IMG_DIR,
label_dir=config.LABEL_DIR,
anchors=config.ANCHORS,
)
if config.LOAD_MODEL:
load_checkpoint(
config.CHECKPOINT_FILE, self.model, self.optimizer, config.LEARNING_RATE)
def setup(self, stage=None):
pass
def train_dataloader(self):
return DataLoader(
dataset=self.train_dataset,
batch_size=config.BATCH_SIZE,
num_workers=config.NUM_WORKERS,
pin_memory=config.PIN_MEMORY,
persistent_workers=True,
shuffle=True,
drop_last=False,
)
def val_dataloader(self):
return DataLoader(
dataset=self.val_dataset,
batch_size=config.BATCH_SIZE,
num_workers=config.NUM_WORKERS,
pin_memory=config.PIN_MEMORY,
persistent_workers=True,
shuffle=False,
drop_last=False,
)
def test_dataloader(self):
return DataLoader(
dataset=self.test_dataset,
batch_size=config.BATCH_SIZE,
num_workers=config.NUM_WORKERS,
pin_memory=config.PIN_MEMORY,
persistent_workers=True,
shuffle=False,
drop_last=False,
)
|