File size: 10,071 Bytes
09a3fa9 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 | # 迁移 MMCV 参数调度器到 MMEngine
MMCV 1.x 版本使用 [LrUpdaterHook](https://mmcv.readthedocs.io/zh_CN/v1.6.0/api.html#mmcv.runner.LrUpdaterHook) 和 [MomentumUpdaterHook](https://mmcv.readthedocs.io/zh_CN/v1.6.0/api.html#mmcv.runner.MomentumUpdaterHook) 来调整学习率和动量。
但随着深度学习算法训练方式的不断发展,使用 Hook 修改学习率已经难以满足更加丰富的自定义需求,因此 MMEngine 提供了参数调度器(ParamScheduler)。
一方面,参数调度器的接口与 PyTroch 的学习率调度器(LRScheduler)对齐,另一方面,参数调度器提供了更丰富的功能,详细请参考[参数调度器使用指南](../tutorials/param_scheduler.md)。
## 学习率调度器(LrUpdater)迁移
MMEngine 中使用 LRScheduler 替代 LrUpdaterHook,配置文件中的字段从原本的 `lr_config` 修改为 `param_scheduler`。
MMCV 中的学习率配置与 MMEngine 中的参数调度器配置对应关系如下:
### 学习率预热(Warmup)迁移
由于 MMEngine 中的学习率调度器在实现时增加了 begin 和 end 参数,指定了调度器的生效区间,所以可以通过调度器组合的方式实现学习率预热。MMCV 中有 3 种学习率预热方式,分别是 `'constant'`, `'linear'`, `'exp'`,在 MMEngine 中对应的配置应修改为:
#### 常数预热(constant)
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
warmup='constant',
warmup_ratio=0.1,
warmup_iters=500,
warmup_by_epoch=False
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='ConstantLR',
factor=0.1,
begin=0,
end=500,
by_epoch=False),
dict(...) # 主学习率调度器配置
]
```
</td>
</tr>
</thead>
</table>
#### 线性预热(linear)
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
warmup='linear',
warmup_ratio=0.1,
warmup_iters=500,
warmup_by_epoch=False
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='LinearLR',
start_factor=0.1,
begin=0,
end=500,
by_epoch=False),
dict(...) # 主学习率调度器配置
]
```
</td>
</tr>
</thead>
</table>
#### 指数预热(exp)
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
warmup='exp',
warmup_ratio=0.1,
warmup_iters=500,
warmup_by_epoch=False
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='ExponentialLR',
gamma=0.1,
begin=0,
end=500,
by_epoch=False),
dict(...) # 主学习率调度器配置
]
```
</td>
</tr>
</thead>
</table>
### fixed 学习率(FixedLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(policy='fixed')
```
</td>
<td>
```python
param_scheduler = [
dict(type='ConstantLR', factor=1)
]
```
</td>
</tr>
</thead>
</table>
### step 学习率(StepLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='step',
step=[8, 11],
gamma=0.1,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='MultiStepLR',
milestone=[8, 11],
gamma=0.1,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### poly 学习率(PolyLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='poly',
power=0.7,
min_lr=0.001,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='PolyLR',
power=0.7,
eta_min=0.001,
begin=0,
end=num_epochs,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### exp 学习率(ExpLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='exp',
power=0.5,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='ExponentialLR',
gamma=0.5,
begin=0,
end=num_epochs,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### CosineAnnealing 学习率(CosineAnnealingLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='CosineAnnealing',
min_lr=0.5,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='CosineAnnealingLR',
eta_min=0.5,
T_max=num_epochs,
begin=0,
end=num_epochs,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### FlatCosineAnnealing 学习率(FlatCosineAnnealingLrUpdaterHook)迁移
像 FlatCosineAnnealing 这种由多个学习率策略拼接而成的学习率,原本需要重写 Hook 来实现,而在 MMEngine 中只需将两个参数调度器组合即可
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='FlatCosineAnnealing',
start_percent=0.5,
min_lr=0.005,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='ConstantLR', factor=1, begin=0, end=num_epochs * 0.75)
dict(type='CosineAnnealingLR',
eta_min=0.005,
begin=num_epochs * 0.75,
end=num_epochs,
T_max=num_epochs * 0.25,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### CosineRestart 学习率(CosineRestartLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(policy='CosineRestart',
periods=[5, 10, 15],
restart_weights=[1, 0.7, 0.3],
min_lr=0.001,
by_epoch=True)
```
</td>
<td>
```python
param_scheduler = [
dict(type='CosineRestartLR',
periods=[5, 10, 15],
restart_weights=[1, 0.7, 0.3],
eta_min=0.001,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
### OneCycle 学习率(OneCycleLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(policy='OneCycle',
max_lr=0.02,
total_steps=90000,
pct_start=0.3,
anneal_strategy='cos',
div_factor=25,
final_div_factor=1e4,
three_phase=True,
by_epoch=False)
```
</td>
<td>
```python
param_scheduler = [
dict(type='OneCycleLR',
eta_max=0.02,
total_steps=90000,
pct_start=0.3,
anneal_strategy='cos',
div_factor=25,
final_div_factor=1e4,
three_phase=True,
by_epoch=False)
]
```
</td>
</tr>
</thead>
</table>
需要注意的是 `by_epoch` 参数 MMCV 默认是 `False`, MMEngine 默认是 `True`
### LinearAnnealing 学习率(LinearAnnealingLrUpdaterHook)迁移
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='LinearAnnealing',
min_lr_ratio=0.01,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
dict(type='LinearLR',
start_factor=1,
end_factor=0.01,
begin=0,
end=num_epochs,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
## 动量调度器(MomentumUpdater)迁移
MMCV 使用 `momentum_config` 字段和 MomentumUpdateHook 调整动量。 MMEngine 中动量同样由参数调度器控制。用户可以简单将学习率调度器后的 `LR` 修改为 `Momentum`,即可使用同样的策略来调整动量。动量调度器只需要和学习率调度器一样添加进 `param_scheduler` 列表中即可。举一个简单的例子:
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(...)
momentum_config = dict(
policy='CosineAnnealing',
min_momentum=0.1,
by_epoch=True
)
```
</td>
<td>
```python
param_scheduler = [
# 学习率调度器配置
dict(...),
# 动量调度器配置
dict(type='CosineAnnealingMomentum',
eta_min=0.1,
T_max=num_epochs,
begin=0,
end=num_epochs,
by_epoch=True)
]
```
</td>
</tr>
</thead>
</table>
## 参数更新频率相关配置迁移
如果在使用 epoch-based 训练循环且配置文件中按 epoch 设置生效区间(`begin`,`end`)或周期(`T_max`)等变量的同时希望参数率按 iteration 更新,在 MMCV 中需要将 `by_epoch` 设置为 False。而在 MMEngine 中需要注意,配置中的 `by_epoch` 仍需设置为 True,通过在配置中添加 `convert_to_iter_based=True` 来构建按 iteration 更新的参数调度器,关于此配置详见[参数调度器教程](../tutorials/param_scheduler.md)。
以迁移CosineAnnealing为例:
<table class="docutils">
<thead>
<tr>
<th>MMCV-1.x</th>
<th>MMEngine</th>
<tbody>
<tr>
<td>
```python
lr_config = dict(
policy='CosineAnnealing',
min_lr=0.5,
by_epoch=False
)
```
</td>
<td>
```python
param_scheduler = [
dict(
type='CosineAnnealingLR',
eta_min=0.5,
T_max=num_epochs,
by_epoch=True, # 注意,by_epoch 需要设置为 True
convert_to_iter_based=True # 转换为按 iter 更新参数
)
]
```
</td>
</tr>
</thead>
</table>
你可能还想阅读[参数调度器的教程](../tutorials/param_scheduler.md)或者[参数调度器的 API 文档](mmengine.optim.scheduler)。
|