From e26182e6ec862870c339d01f4eb849ab5b0c26c7 Mon Sep 17 00:00:00 2001
From: liaoxingyu <sherlockliao01@gmail.com>
Date: Fri, 22 Jan 2021 11:17:21 +0800
Subject: [PATCH] make lr warmup by iter
Summary: change warmup way by iter not by epoch, which will make it more flexible when training small epochs
---
docs/GETTING_STARTED.md => GETTING_STARTED.md | 0
docs/INSTALL.md => INSTALL.md | 0
docs/MODEL_ZOO.md => MODEL_ZOO.md | 0
README.md | 4 ++--
configs/Base-SBS.yml | 2 +-
configs/Base-bagtricks.yml | 2 +-
configs/VERIWild/bagtricks_R50-ibn.yml | 2 +-
configs/VeRi/sbs_R50-ibn.yml | 2 +-
configs/VehicleID/bagtricks_R50-ibn.yml | 2 +-
fastreid/__init__.py | 2 +-
fastreid/config/defaults.py | 2 +-
fastreid/engine/defaults.py | 10 +++++-----
fastreid/engine/hooks.py | 11 +++++++----
fastreid/modeling/backbones/regnet/config.py | 2 +-
.../regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-12GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-16GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-3.2GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-32GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-4.0GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-6.4GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-600MF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-8.0GF_dds_8gpu.yaml | 2 +-
.../regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-1.6GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-12GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-16GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-200MF_dds_8gpu.yaml | 1 -
.../regnet/regnety/RegNetY-3.2GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-32GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-4.0GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-400MF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-6.4GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-600MF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-8.0GF_dds_8gpu.yaml | 2 +-
.../regnet/regnety/RegNetY-800MF_dds_8gpu.yaml | 2 +-
fastreid/modeling/meta_arch/distiller.py | 2 +-
fastreid/solver/build.py | 16 ++++++++--------
fastreid/solver/lr_scheduler.py | 18 +++++++++---------
fastreid/utils/checkpoint.py | 14 ++++++++++++++
projects/FastAttr/configs/Base-attribute.yml | 2 +-
projects/FastCls/configs/base-cls.yaml | 2 +-
projects/FastFace/configs/face_base.yml | 2 +-
.../FastRetri/configs/base-image_retri.yml | 2 +-
projects/FastTune/configs/search_trial.yml | 2 +-
.../PartialReID/configs/partial_market.yml | 2 +-
docs/requirements => requirements.txt | 2 +-
tools/plain_train_net.py | 14 +++++++-------
50 files changed, 91 insertions(+), 75 deletions(-)
rename docs/GETTING_STARTED.md => GETTING_STARTED.md (100%)
rename docs/INSTALL.md => INSTALL.md (100%)
rename docs/MODEL_ZOO.md => MODEL_ZOO.md (100%)
rename docs/requirements => requirements.txt (95%)
diff --git a/docs/GETTING_STARTED.md b/GETTING_STARTED.md
similarity index 100%
rename from docs/GETTING_STARTED.md
rename to GETTING_STARTED.md
diff --git a/docs/INSTALL.md b/INSTALL.md
similarity index 100%
rename from docs/INSTALL.md
rename to INSTALL.md
diff --git a/docs/MODEL_ZOO.md b/MODEL_ZOO.md
similarity index 100%
rename from docs/MODEL_ZOO.md
rename to MODEL_ZOO.md
diff --git a/README.md b/README.md
index 0203e52..e8e64cf 100644
--- a/README.md
+++ b/README.md
@@ -26,13 +26,13 @@ See [INSTALL.md](https://github.com/JDAI-CV/fast-reid/blob/master/docs/INSTALL.m
The designed architecture follows this guide [PyTorch-Project-Template](https://github.com/L1aoXingyu/PyTorch-Project-Template), you can check each folder's purpose by yourself.
-See [GETTING_STARTED.md](https://github.com/JDAI-CV/fast-reid/blob/master/docs/GETTING_STARTED.md).
+See [GETTING_STARTED.md](https://github.com/JDAI-CV/fast-reid/blob/master/GETTING_STARTED.md).
Learn more at out [documentation](). And see [projects/](https://github.com/JDAI-CV/fast-reid/tree/master/projects) for some projects that are build on top of fastreid.
## Model Zoo and Baselines
-We provide a large set of baseline results and trained models available for download in the [Fastreid Model Zoo](https://github.com/JDAI-CV/fast-reid/blob/master/docs/MODEL_ZOO.md).
+We provide a large set of baseline results and trained models available for download in the [Fastreid Model Zoo](https://github.com/JDAI-CV/fast-reid/blob/master/MODEL_ZOO.md).
## Deployment
diff --git a/configs/Base-SBS.yml b/configs/Base-SBS.yml
index 04d1f9b..5085a59 100644
--- a/configs/Base-SBS.yml
+++ b/configs/Base-SBS.yml
@@ -50,7 +50,7 @@ SOLVER:
ETA_MIN_LR: 0.0000007
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
FREEZE_ITERS: 1000
diff --git a/configs/Base-bagtricks.yml b/configs/Base-bagtricks.yml
index a3da846..204346f 100644
--- a/configs/Base-bagtricks.yml
+++ b/configs/Base-bagtricks.yml
@@ -60,7 +60,7 @@ SOLVER:
GAMMA: 0.1
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
CHECKPOINT_PERIOD: 30
diff --git a/configs/VERIWild/bagtricks_R50-ibn.yml b/configs/VERIWild/bagtricks_R50-ibn.yml
index 77a52c2..de6da4a 100644
--- a/configs/VERIWild/bagtricks_R50-ibn.yml
+++ b/configs/VERIWild/bagtricks_R50-ibn.yml
@@ -22,7 +22,7 @@ SOLVER:
IMS_PER_BATCH: 128
MAX_ITER: 60
STEPS: [30, 50]
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
CHECKPOINT_PERIOD: 20
diff --git a/configs/VeRi/sbs_R50-ibn.yml b/configs/VeRi/sbs_R50-ibn.yml
index 724d460..686b1fc 100644
--- a/configs/VeRi/sbs_R50-ibn.yml
+++ b/configs/VeRi/sbs_R50-ibn.yml
@@ -16,7 +16,7 @@ SOLVER:
IMS_PER_BATCH: 64
MAX_ITER: 60
DELAY_ITERS: 30
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
FREEZE_ITERS: 10
CHECKPOINT_PERIOD: 20
diff --git a/configs/VehicleID/bagtricks_R50-ibn.yml b/configs/VehicleID/bagtricks_R50-ibn.yml
index 9ce8456..4b1c439 100644
--- a/configs/VehicleID/bagtricks_R50-ibn.yml
+++ b/configs/VehicleID/bagtricks_R50-ibn.yml
@@ -24,7 +24,7 @@ SOLVER:
IMS_PER_BATCH: 512
MAX_ITER: 60
STEPS: [30, 50]
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
CHECKPOINT_PERIOD: 20
diff --git a/fastreid/__init__.py b/fastreid/__init__.py
index 94eab6e..0efd06c 100644
--- a/fastreid/__init__.py
+++ b/fastreid/__init__.py
@@ -5,4 +5,4 @@
"""
-__version__ = "0.2.0"
+__version__ = "1.0.0"
diff --git a/fastreid/config/defaults.py b/fastreid/config/defaults.py
index 3ac8251..dbc2170 100644
--- a/fastreid/config/defaults.py
+++ b/fastreid/config/defaults.py
@@ -238,7 +238,7 @@ _C.SOLVER.ETA_MIN_LR = 1e-7
# Warmup options
_C.SOLVER.WARMUP_FACTOR = 0.1
-_C.SOLVER.WARMUP_EPOCHS = 10
+_C.SOLVER.WARMUP_ITERS = 1000
_C.SOLVER.WARMUP_METHOD = "linear"
# Backbone freeze iters
diff --git a/fastreid/engine/defaults.py b/fastreid/engine/defaults.py
index 3d52bfa..28bb560 100644
--- a/fastreid/engine/defaults.py
+++ b/fastreid/engine/defaults.py
@@ -233,7 +233,8 @@ class DefaultTrainer(TrainerBase):
model, data_loader, optimizer
)
- self.scheduler = self.build_lr_scheduler(cfg, optimizer)
+ self.iters_per_epoch = len(data_loader.dataset) // cfg.SOLVER.IMS_PER_BATCH
+ self.scheduler = self.build_lr_scheduler(cfg, optimizer, self.iters_per_epoch)
# Assume no other objects need to be checkpointed.
# We can later make it checkpoint the stateful hooks
@@ -246,12 +247,11 @@ class DefaultTrainer(TrainerBase):
**self.scheduler,
)
- self.iters_per_epoch = len(data_loader.dataset) // cfg.SOLVER.IMS_PER_BATCH
self.start_epoch = 0
self.max_epoch = cfg.SOLVER.MAX_EPOCH
self.max_iter = self.max_epoch * self.iters_per_epoch
- self.warmup_epochs = cfg.SOLVER.WARMUP_EPOCHS
+ self.warmup_iters = cfg.SOLVER.WARMUP_ITERS
self.delay_epochs = cfg.SOLVER.DELAY_EPOCHS
self.cfg = cfg
@@ -409,12 +409,12 @@ class DefaultTrainer(TrainerBase):
return build_optimizer(cfg, model)
@classmethod
- def build_lr_scheduler(cls, cfg, optimizer):
+ def build_lr_scheduler(cls, cfg, optimizer, iters_per_epoch):
"""
It now calls :func:`fastreid.solver.build_lr_scheduler`.
Overwrite it if you'd like a different scheduler.
"""
- return build_lr_scheduler(cfg, optimizer)
+ return build_lr_scheduler(cfg, optimizer, iters_per_epoch)
@classmethod
def build_train_loader(cls, cfg):
diff --git a/fastreid/engine/hooks.py b/fastreid/engine/hooks.py
index 91fbaf9..f7111a3 100644
--- a/fastreid/engine/hooks.py
+++ b/fastreid/engine/hooks.py
@@ -250,11 +250,14 @@ class LRScheduler(HookBase):
lr = self._optimizer.param_groups[self._best_param_group_id]["lr"]
self.trainer.storage.put_scalar("lr", lr, smoothing_hint=False)
- def after_epoch(self):
- next_epoch = self.trainer.epoch + 1
- if next_epoch <= self.trainer.warmup_epochs:
+ next_iter = self.trainer.iter + 1
+ if next_iter <= self.trainer.warmup_iters:
self._scheduler["warmup_sched"].step()
- elif next_epoch >= self.trainer.delay_epochs:
+
+ def after_epoch(self):
+ next_iter = self.trainer.iter + 1
+ next_epoch = self.trainer.epoch + 1
+ if next_iter > self.trainer.warmup_iters and next_epoch >= self.trainer.delay_epochs:
self._scheduler["lr_sched"].step()
diff --git a/fastreid/modeling/backbones/regnet/config.py b/fastreid/modeling/backbones/regnet/config.py
index 4496480..cb967dc 100644
--- a/fastreid/modeling/backbones/regnet/config.py
+++ b/fastreid/modeling/backbones/regnet/config.py
@@ -224,7 +224,7 @@ _C.OPTIM.WEIGHT_DECAY = 5e-4
_C.OPTIM.WARMUP_FACTOR = 0.1
# Gradually warm up the OPTIM.BASE_LR over this number of epochs
-_C.OPTIM.WARMUP_EPOCHS = 0
+_C.OPTIM.WARMUP_ITERS = 0
# ------------------------------------------------------------------------------------ #
# Training options
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml
index c8133d7..7127fd1 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-1.6GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-12GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-12GF_dds_8gpu.yaml
index a3edf42..63d4ef9 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-12GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-12GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-16GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-16GF_dds_8gpu.yaml
index 0f94f9b..768763b 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-16GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-16GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml
index c0b13b6..e41c967 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-200MF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-3.2GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-3.2GF_dds_8gpu.yaml
index 594d533..f7af0d7 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-3.2GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-3.2GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-32GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-32GF_dds_8gpu.yaml
index c1d34b8..9dfb78b 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-32GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-32GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-4.0GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-4.0GF_dds_8gpu.yaml
index bd95453..d884ed6 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-4.0GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-4.0GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml
index 7b887ad..1c14a16 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-400MF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-6.4GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-6.4GF_dds_8gpu.yaml
index f256e64..48aeb83 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-6.4GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-6.4GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-600MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-600MF_dds_8gpu.yaml
index aca28aa..8ddf668 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-600MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-600MF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-8.0GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-8.0GF_dds_8gpu.yaml
index a4141d6..b75beb0 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-8.0GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-8.0GF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml
index 8d2f6ae..8e76db7 100644
--- a/fastreid/modeling/backbones/regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnetx/RegNetX-800MF_dds_8gpu.yaml
@@ -13,7 +13,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-1.6GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-1.6GF_dds_8gpu.yaml
index 2dc9f37..25c76e9 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-1.6GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-1.6GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-12GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-12GF_dds_8gpu.yaml
index 6d27d5d..5dea853 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-12GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-12GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-16GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-16GF_dds_8gpu.yaml
index 605d215..9d77d4c 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-16GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-16GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-200MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-200MF_dds_8gpu.yaml
index 300cc43..cfe006b 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-200MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-200MF_dds_8gpu.yaml
@@ -14,7 +14,6 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-3.2GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-3.2GF_dds_8gpu.yaml
index 95f05ba..70f5f37 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-3.2GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-3.2GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-32GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-32GF_dds_8gpu.yaml
index 753d7a5..89e83e1 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-32GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-32GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-4.0GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-4.0GF_dds_8gpu.yaml
index 27895a9..4a1afab 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-4.0GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-4.0GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-400MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-400MF_dds_8gpu.yaml
index 1b1c31b..0cca887 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-400MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-400MF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-6.4GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-6.4GF_dds_8gpu.yaml
index 74535c2..a93c4ae 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-6.4GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-6.4GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-600MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-600MF_dds_8gpu.yaml
index 661e1a9..adc490d 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-600MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-600MF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-8.0GF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-8.0GF_dds_8gpu.yaml
index 792147a..78377ed 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-8.0GF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-8.0GF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/backbones/regnet/regnety/RegNetY-800MF_dds_8gpu.yaml b/fastreid/modeling/backbones/regnet/regnety/RegNetY-800MF_dds_8gpu.yaml
index 6e52823..c6db355 100644
--- a/fastreid/modeling/backbones/regnet/regnety/RegNetY-800MF_dds_8gpu.yaml
+++ b/fastreid/modeling/backbones/regnet/regnety/RegNetY-800MF_dds_8gpu.yaml
@@ -14,7 +14,7 @@ OPTIM:
MAX_EPOCH: 100
MOMENTUM: 0.9
WEIGHT_DECAY: 5e-5
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 5
TRAIN:
DATASET: imagenet
IM_SIZE: 224
diff --git a/fastreid/modeling/meta_arch/distiller.py b/fastreid/modeling/meta_arch/distiller.py
index 1ee3008..5405721 100644
--- a/fastreid/modeling/meta_arch/distiller.py
+++ b/fastreid/modeling/meta_arch/distiller.py
@@ -64,7 +64,7 @@ class Distiller(Baseline):
return super(Distiller, self).forward(batched_inputs)
def losses(self, s_outputs, t_outputs, gt_labels):
- r"""
+ """
Compute loss from modeling's outputs, the loss function input arguments
must be the same as the outputs of the model forwarding.
"""
diff --git a/fastreid/solver/build.py b/fastreid/solver/build.py
index 524a72e..806aa7e 100644
--- a/fastreid/solver/build.py
+++ b/fastreid/solver/build.py
@@ -4,6 +4,8 @@
@contact: sherlockliao01@gmail.com
"""
+import math
+
from . import lr_scheduler
from . import optim
@@ -34,11 +36,9 @@ def build_optimizer(cfg, model):
return opt_fns
-def build_lr_scheduler(cfg, optimizer):
- cfg = cfg.clone()
- cfg.defrost()
- cfg.SOLVER.MAX_EPOCH = cfg.SOLVER.MAX_EPOCH - max(
- cfg.SOLVER.WARMUP_EPOCHS + 1, cfg.SOLVER.DELAY_EPOCHS)
+def build_lr_scheduler(cfg, optimizer, iters_per_epoch):
+ max_epoch = cfg.SOLVER.MAX_EPOCH - max(
+ math.ceil(cfg.SOLVER.WARMUP_ITERS / iters_per_epoch), cfg.SOLVER.DELAY_EPOCHS)
scheduler_dict = {}
@@ -52,7 +52,7 @@ def build_lr_scheduler(cfg, optimizer):
"CosineAnnealingLR": {
"optimizer": optimizer,
# cosine annealing lr scheduler options
- "T_max": cfg.SOLVER.MAX_EPOCH,
+ "T_max": max_epoch,
"eta_min": cfg.SOLVER.ETA_MIN_LR,
},
@@ -61,13 +61,13 @@ def build_lr_scheduler(cfg, optimizer):
scheduler_dict["lr_sched"] = getattr(lr_scheduler, cfg.SOLVER.SCHED)(
**scheduler_args[cfg.SOLVER.SCHED])
- if cfg.SOLVER.WARMUP_EPOCHS > 0:
+ if cfg.SOLVER.WARMUP_ITERS > 0:
warmup_args = {
"optimizer": optimizer,
# warmup options
"warmup_factor": cfg.SOLVER.WARMUP_FACTOR,
- "warmup_epochs": cfg.SOLVER.WARMUP_EPOCHS,
+ "warmup_iters": cfg.SOLVER.WARMUP_ITERS,
"warmup_method": cfg.SOLVER.WARMUP_METHOD,
}
scheduler_dict["warmup_sched"] = lr_scheduler.WarmupLR(**warmup_args)
diff --git a/fastreid/solver/lr_scheduler.py b/fastreid/solver/lr_scheduler.py
index 711c1b5..17ce2b9 100644
--- a/fastreid/solver/lr_scheduler.py
+++ b/fastreid/solver/lr_scheduler.py
@@ -15,18 +15,18 @@ class WarmupLR(torch.optim.lr_scheduler._LRScheduler):
self,
optimizer: torch.optim.Optimizer,
warmup_factor: float = 0.1,
- warmup_epochs: int = 10,
+ warmup_iters: int = 1000,
warmup_method: str = "linear",
last_epoch: int = -1,
):
self.warmup_factor = warmup_factor
- self.warmup_epochs = warmup_epochs
+ self.warmup_iters = warmup_iters
self.warmup_method = warmup_method
super().__init__(optimizer, last_epoch)
def get_lr(self) -> List[float]:
warmup_factor = _get_warmup_factor_at_epoch(
- self.warmup_method, self.last_epoch, self.warmup_epochs, self.warmup_factor
+ self.warmup_method, self.last_epoch, self.warmup_iters, self.warmup_factor
)
return [
base_lr * warmup_factor for base_lr in self.base_lrs
@@ -38,29 +38,29 @@ class WarmupLR(torch.optim.lr_scheduler._LRScheduler):
def _get_warmup_factor_at_epoch(
- method: str, epoch: int, warmup_epochs: int, warmup_factor: float
+ method: str, iter: int, warmup_iters: int, warmup_factor: float
) -> float:
"""
Return the learning rate warmup factor at a specific iteration.
See https://arxiv.org/abs/1706.02677 for more details.
Args:
method (str): warmup method; either "constant" or "linear".
- epoch (int): epoch at which to calculate the warmup factor.
- warmup_epochs (int): the number of warmup epochs.
+ iter (int): iter at which to calculate the warmup factor.
+ warmup_iters (int): the number of warmup epochs.
warmup_factor (float): the base warmup factor (the meaning changes according
to the method used).
Returns:
float: the effective warmup factor at the given iteration.
"""
- if epoch >= warmup_epochs:
+ if iter >= warmup_iters:
return 1.0
if method == "constant":
return warmup_factor
elif method == "linear":
- alpha = epoch / warmup_epochs
+ alpha = iter / warmup_iters
return warmup_factor * (1 - alpha) + alpha
elif method == "exp":
- return warmup_factor ** (1 - epoch / warmup_epochs)
+ return warmup_factor ** (1 - iter / warmup_iters)
else:
raise ValueError("Unknown warmup method: {}".format(method))
diff --git a/fastreid/utils/checkpoint.py b/fastreid/utils/checkpoint.py
index 1f142d0..b31c4c0 100644
--- a/fastreid/utils/checkpoint.py
+++ b/fastreid/utils/checkpoint.py
@@ -73,6 +73,7 @@ class Checkpointer(object):
def save(self, name: str, **kwargs: Dict[str, str]):
"""
Dump model and checkpointables to a file.
+
Args:
name (str): name of the file.
kwargs (dict): extra arbitrary data to save.
@@ -98,6 +99,7 @@ class Checkpointer(object):
"""
Load from the given checkpoint. When path points to network file, this
function has to be called on all ranks.
+
Args:
path (str): path or url to the checkpoint. If empty, will not load
anything.
@@ -176,6 +178,7 @@ class Checkpointer(object):
If `resume` is True, this method attempts to resume from the last
checkpoint, if exists. Otherwise, load checkpoint from the given path.
This is useful when restarting an interrupted training job.
+
Args:
path (str): path to the checkpoint.
resume (bool): if True, resume from the last checkpoint if it exists.
@@ -191,6 +194,7 @@ class Checkpointer(object):
def tag_last_checkpoint(self, last_filename_basename: str):
"""
Tag the last checkpoint.
+
Args:
last_filename_basename (str): the basename of the last filename.
"""
@@ -202,6 +206,7 @@ class Checkpointer(object):
"""
Load a checkpoint file. Can be overwritten by subclasses to support
different formats.
+
Args:
f (str): a locally mounted file path.
Returns:
@@ -214,6 +219,7 @@ class Checkpointer(object):
def _load_model(self, checkpoint: Any):
"""
Load weights from a checkpoint.
+
Args:
checkpoint (Any): checkpoint contains the weights.
"""
@@ -269,6 +275,7 @@ class Checkpointer(object):
def _convert_ndarray_to_tensor(self, state_dict: dict):
"""
In-place convert all numpy arrays in the state_dict to torch tensor.
+
Args:
state_dict (dict): a state-dict to be loaded to the model.
"""
@@ -313,6 +320,7 @@ class PeriodicCheckpointer:
def step(self, epoch: int, **kwargs: Any):
"""
Perform the appropriate action at the given iteration.
+
Args:
epoch (int): the current epoch, ranged in [0, max_epoch-1].
kwargs (Any): extra data to save, same as in
@@ -342,6 +350,7 @@ class PeriodicCheckpointer:
"""
Same argument as :meth:`Checkpointer.save`.
Use this method to manually save checkpoints outside the schedule.
+
Args:
name (str): file name.
kwargs (Any): extra data to save, same as in
@@ -374,6 +383,7 @@ def get_missing_parameters_message(keys: List[str]) -> str:
"""
Get a logging-friendly message to report parameter names (keys) that are in
the model but not found in a checkpoint.
+
Args:
keys (list[str]): List of keys that were not found in the checkpoint.
Returns:
@@ -391,6 +401,7 @@ def get_unexpected_parameters_message(keys: List[str]) -> str:
"""
Get a logging-friendly message to report parameter names (keys) that are in
the checkpoint but not found in the model.
+
Args:
keys (list[str]): List of keys that were not found in the model.
Returns:
@@ -407,6 +418,7 @@ def get_unexpected_parameters_message(keys: List[str]) -> str:
def _strip_prefix_if_present(state_dict: Dict[str, Any], prefix: str) -> None:
"""
Strip the prefix in metadata, if any.
+
Args:
state_dict (OrderedDict): a state-dict to be loaded to the model.
prefix (str): prefix.
@@ -441,6 +453,7 @@ def _group_checkpoint_keys(keys: List[str]) -> Dict[str, List[str]]:
"""
Group keys based on common prefixes. A prefix is the string up to the final
"." in each key.
+
Args:
keys (list[str]): list of parameter names, i.e. keys in the model
checkpoint dict.
@@ -461,6 +474,7 @@ def _group_checkpoint_keys(keys: List[str]) -> Dict[str, List[str]]:
def _group_to_str(group: List[str]) -> str:
"""
Format a group of parameter name suffixes into a loggable string.
+
Args:
group (list[str]): list of parameter name suffixes.
Returns:
diff --git a/projects/FastAttr/configs/Base-attribute.yml b/projects/FastAttr/configs/Base-attribute.yml
index 33eb452..f9ee16e 100644
--- a/projects/FastAttr/configs/Base-attribute.yml
+++ b/projects/FastAttr/configs/Base-attribute.yml
@@ -50,7 +50,7 @@ SOLVER:
STEPS: [ 15, 20, 25 ]
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 0
+ WARMUP_ITERS: 1000
CHECKPOINT_PERIOD: 10
diff --git a/projects/FastCls/configs/base-cls.yaml b/projects/FastCls/configs/base-cls.yaml
index d02747a..c355703 100644
--- a/projects/FastCls/configs/base-cls.yaml
+++ b/projects/FastCls/configs/base-cls.yaml
@@ -60,7 +60,7 @@ SOLVER:
ETA_MIN_LR: 0.00003
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 2000
CHECKPOINT_PERIOD: 10
diff --git a/projects/FastFace/configs/face_base.yml b/projects/FastFace/configs/face_base.yml
index c6e94d5..8a5f5f8 100644
--- a/projects/FastFace/configs/face_base.yml
+++ b/projects/FastFace/configs/face_base.yml
@@ -59,7 +59,7 @@ SOLVER:
IMS_PER_BATCH: 512
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 1
+ WARMUP_ITERS: 5000
CHECKPOINT_PERIOD: 2
diff --git a/projects/FastRetri/configs/base-image_retri.yml b/projects/FastRetri/configs/base-image_retri.yml
index 85df78d..d1bfa87 100644
--- a/projects/FastRetri/configs/base-image_retri.yml
+++ b/projects/FastRetri/configs/base-image_retri.yml
@@ -61,7 +61,7 @@ SOLVER:
ETA_MIN_LR: 0.00003
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 10
+ WARMUP_ITERS: 1000
CHECKPOINT_PERIOD: 10
diff --git a/projects/FastTune/configs/search_trial.yml b/projects/FastTune/configs/search_trial.yml
index 1220dfe..18ae173 100644
--- a/projects/FastTune/configs/search_trial.yml
+++ b/projects/FastTune/configs/search_trial.yml
@@ -71,7 +71,7 @@ SOLVER:
FREEZE_ITERS: 500
WARMUP_FACTOR: 0.1
- WARMUP_EPOCHS: 5
+ WARMUP_ITERS: 1000
CHECKPOINT_PERIOD: 100
diff --git a/projects/PartialReID/configs/partial_market.yml b/projects/PartialReID/configs/partial_market.yml
index b78706e..d43a4e5 100644
--- a/projects/PartialReID/configs/partial_market.yml
+++ b/projects/PartialReID/configs/partial_market.yml
@@ -57,7 +57,7 @@ SOLVER:
GAMMA: 0.1
WARMUP_FACTOR: 0.01
- WARMUP_ITERS: 5
+ WARMUP_ITERS: 1000
CHECKPOINT_PERIOD: 10
diff --git a/docs/requirements b/requirements.txt
similarity index 95%
rename from docs/requirements
rename to requirements.txt
index f9f6594..a53f22c 100644
--- a/docs/requirements
+++ b/requirements.txt
@@ -17,4 +17,4 @@ termcolor
scikit-learn
tabulate
gdown
-faiss-cpu
\ No newline at end of file
+faiss-gpu
\ No newline at end of file
diff --git a/tools/plain_train_net.py b/tools/plain_train_net.py
index 2d62360..53ce9c3 100644
--- a/tools/plain_train_net.py
+++ b/tools/plain_train_net.py
@@ -80,7 +80,8 @@ def do_train(cfg, model, resume=False):
optimizer_ckpt = dict(optimizer=optimizer)
- scheduler = build_lr_scheduler(cfg, optimizer)
+ iters_per_epoch = len(data_loader.dataset) // cfg.SOLVER.IMS_PER_BATCH
+ scheduler = build_lr_scheduler(cfg, optimizer, iters_per_epoch)
checkpointer = Checkpointer(
model,
@@ -90,8 +91,6 @@ def do_train(cfg, model, resume=False):
**scheduler
)
- iters_per_epoch = len(data_loader.dataset) // cfg.SOLVER.IMS_PER_BATCH
-
start_epoch = (
checkpointer.resume_or_load(cfg.MODEL.WEIGHTS, resume=resume).get("epoch", -1) + 1
)
@@ -99,7 +98,7 @@ def do_train(cfg, model, resume=False):
max_epoch = cfg.SOLVER.MAX_EPOCH
max_iter = max_epoch * iters_per_epoch
- warmup_epochs = cfg.SOLVER.WARMUP_EPOCHS
+ warmup_iters = cfg.SOLVER.WARMUP_ITERS
delay_epochs = cfg.SOLVER.DELAY_EPOCHS
periodic_checkpointer = PeriodicCheckpointer(checkpointer, cfg.SOLVER.CHECKPOINT_PERIOD, max_epoch)
@@ -146,13 +145,14 @@ def do_train(cfg, model, resume=False):
iteration += 1
+ if iteration <= warmup_iters:
+ scheduler["warmup_sched"].step()
+
# Write metrics after each epoch
for writer in writers:
writer.write()
- if (epoch + 1) <= warmup_epochs:
- scheduler["warmup_sched"].step()
- elif (epoch + 1) >= delay_epochs:
+ if iteration > warmup_iters and (epoch + 1) >= delay_epochs:
scheduler["lr_sched"].step()
if (