update ssl configs and readthedocs

README.md

@@ -109,7 +109,19 @@ Please see [Tutorials](docs/en/tutorials) for more developing examples and tech
 - [add new modules](docs/en/tutorials/3_new_module.md)
 - [customize schedules](docs/en/tutorials/4_schedule.md)
 - [customize runtime](docs/en/tutorials/5_runtime.md)
-- [benchmarks](docs/en/tutorials/6_benchmarks.md)
+
+Downetream Tasks for Self-supervised Learning
+
+- [Classification](docs/en/tutorials/ssl_classification.md)
+- [Detection](docs/en/tutorials/ssl_detection.md)
+- [Segmentation](docs/en/tutorials/ssl_segmentation.md)
+
+Useful Tools
+
+- [Analysis](docs/en/tutorials/analysis.md)
+- [Visualization](docs/en/tutorials/visualization.md)
+- [pytorch2onnx](docs/en/tutorials/pytorch2onnx.md)
+- [pytorch2torchscript](docs/en/tutorials/pytorch2torchscript.md)
 
 <p align="right">(<a href="#top">back to top</a>)</p>
 

benchmarks/mmdetection/mim_dist_train_full.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+
+set -x
+
+CFG=$1
+PRETRAIN=$2  # pretrained model
+GPUS=$3
+PY_ARGS=${@:4}
+
+# set work_dir according to config path and pretrained model to distinguish different models
+WORK_DIR="$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/$(echo $PRETRAIN | rev | cut -d/ -f 1 | rev)"
+
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+mim train mmdet $CFG \
+    --launcher pytorch -G $GPUS \
+    --work-dir $WORK_DIR \
+    --cfg-options model.backbone.init_cfg.type=Pretrained \
+    model.backbone.init_cfg.checkpoint=$PRETRAIN \
+    model.backbone.init_cfg.prefix="backbone." \
+    $PY_ARGS

benchmarks/mmsegmentation/mim_dist_train.sh

@@ -16,5 +16,4 @@ mim train mmseg $CFG \
     --work-dir $WORK_DIR \
     --cfg-options model.backbone.init_cfg.type=Pretrained \
     model.backbone.init_cfg.checkpoint=$PRETRAIN \
-    model.backbone.init_cfg.prefix="backbone." \
     $PY_ARGS

benchmarks/mmsegmentation/mim_dist_train_full.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+
+set -x
+
+CFG=$1
+PRETRAIN=$2  # pretrained model
+GPUS=$3
+PY_ARGS=${@:4}
+
+# set work_dir according to config path and pretrained model to distinguish different models
+WORK_DIR="$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/$(echo $PRETRAIN | rev | cut -d/ -f 1 | rev)"
+
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+mim train mmseg $CFG \
+    --launcher pytorch -G $GPUS \
+    --work-dir $WORK_DIR \
+    --cfg-options model.backbone.init_cfg.type=Pretrained \
+    model.backbone.init_cfg.checkpoint=$PRETRAIN \
+    model.backbone.init_cfg.prefix="backbone." \
+    $PY_ARGS

benchmarks/mmsegmentation/mim_slurm_train.sh

@@ -24,5 +24,4 @@ mim train mmseg $CFG \
     --srun-args "$SRUN_ARGS" \
     --cfg-options model.backbone.init_cfg.type=Pretrained \
     model.backbone.init_cfg.checkpoint=$PRETRAIN \
-    model.backbone.init_cfg.prefix="backbone." \
     $PY_ARGS

configs/benchmarks/classification/_base_/models/vit_huge_p14.py

@@ -0,0 +1,23 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],  # deit setting
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='MIMVisionTransformer',
+        arch='huge',
+        img_size=224,
+        patch_size=14,
+        drop_path_rate=0.3,
+        final_norm=False,
+        finetune=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',  # mixup CE + label smooth
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,  # no gap in ViT
+        in_channels=1280, num_classes=1000),
+)

configs/benchmarks/classification/imagenet/vit_huge_p14_swin_ft_mae_sz224_8xb128_cos_ep50.py

@@ -0,0 +1,48 @@
+_base_ = [
+    '../_base_/models/vit_huge_p14.py',
+    '../_base_/datasets/imagenet_swin_ft_sz224_8xbs128.py',
+    '../_base_/default_runtime.py',
+]
+
+# data
+data = dict(imgs_per_gpu=128, workers_per_gpu=8)
+
+# interval for accumulate gradient
+update_interval = 1  # total: 8 x bs128 x 1 accumulates = bs1024
+
+# optimizer
+optimizer = dict(
+    type='AdamW',
+    lr=2e-3,
+    weight_decay=0.05, eps=1e-8, betas=(0.9, 0.999),
+    paramwise_options={
+        '(bn|ln|gn)(\d+)?.(weight|bias)': dict(weight_decay=0.),
+        'norm': dict(weight_decay=0.),
+        'bias': dict(weight_decay=0.),
+        'cls_token': dict(weight_decay=0.),
+        'pos_embed': dict(weight_decay=0.),
+    },
+    constructor='TransformerFinetuneConstructor',
+    model_type='vit',
+    layer_decay=0.65)
+
+# learning policy
+lr_config = dict(
+    policy='StepFixCosineAnnealing',
+    min_lr=1e-6,
+    warmup='linear',
+    warmup_iters=5,
+    warmup_ratio=1e-4,
+    warmup_by_epoch=True,
+    by_epoch=False)
+
+# apex
+use_fp16 = True
+fp16 = dict(type='mmcv', loss_scale='dynamic')
+# optimizer args
+optimizer_config = dict(
+    update_interval=update_interval, grad_clip=dict(max_norm=5.0),
+)
+
+# runtime settings
+runner = dict(type='EpochBasedRunner', max_epochs=50)

configs/benchmarks/classification/imagenet/vit_large_p16_swin_ft_mae_sz224_8xb128_cos_ep100.py

@@ -0,0 +1,48 @@
+_base_ = [
+    '../_base_/models/vit_large_p16.py',
+    '../_base_/datasets/imagenet_swin_ft_sz224_8xbs128.py',
+    '../_base_/default_runtime.py',
+]
+
+# data
+data = dict(imgs_per_gpu=128, workers_per_gpu=8)
+
+# interval for accumulate gradient
+update_interval = 1  # total: 8 x bs128 x 1 accumulates = bs1024
+
+# optimizer
+optimizer = dict(
+    type='AdamW',
+    lr=2e-3,
+    weight_decay=0.05, eps=1e-8, betas=(0.9, 0.999),
+    paramwise_options={
+        '(bn|ln|gn)(\d+)?.(weight|bias)': dict(weight_decay=0.),
+        'norm': dict(weight_decay=0.),
+        'bias': dict(weight_decay=0.),
+        'cls_token': dict(weight_decay=0.),
+        'pos_embed': dict(weight_decay=0.),
+    },
+    constructor='TransformerFinetuneConstructor',
+    model_type='vit',
+    layer_decay=0.65)
+
+# learning policy
+lr_config = dict(
+    policy='StepFixCosineAnnealing',
+    min_lr=1e-6,
+    warmup='linear',
+    warmup_iters=5,
+    warmup_ratio=1e-4,
+    warmup_by_epoch=True,
+    by_epoch=False)
+
+# apex
+use_fp16 = True
+fp16 = dict(type='mmcv', loss_scale='dynamic')
+# optimizer args
+optimizer_config = dict(
+    update_interval=update_interval, grad_clip=dict(max_norm=5.0),
+)
+
+# runtime settings
+runner = dict(type='EpochBasedRunner', max_epochs=100)

configs/benchmarks/mmdetection/_base_/datasets/coco_instance.py

@@ -0,0 +1,49 @@
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
+    dict(type='Resize', img_scale=(1333, 800), keep_ratio=True),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(1333, 800),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        pipeline=test_pipeline))
+evaluation = dict(metric=['bbox', 'segm'])

configs/benchmarks/mmdetection/_base_/datasets/voc0712.py

@@ -0,0 +1,55 @@
+# dataset settings
+dataset_type = 'VOCDataset'
+data_root = 'data/VOCdevkit/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='Resize', img_scale=(1000, 600), keep_ratio=True),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(1000, 600),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type='RepeatDataset',
+        times=3,
+        dataset=dict(
+            type=dataset_type,
+            ann_file=[
+                data_root + 'VOC2007/ImageSets/Main/trainval.txt',
+                data_root + 'VOC2012/ImageSets/Main/trainval.txt'
+            ],
+            img_prefix=[data_root + 'VOC2007/', data_root + 'VOC2012/'],
+            pipeline=train_pipeline)),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'VOC2007/ImageSets/Main/test.txt',
+        img_prefix=data_root + 'VOC2007/',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'VOC2007/ImageSets/Main/test.txt',
+        img_prefix=data_root + 'VOC2007/',
+        pipeline=test_pipeline))
+evaluation = dict(interval=1, metric='mAP')

configs/benchmarks/mmdetection/_base_/default_runtime.py

@@ -0,0 +1,27 @@
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+custom_hooks = [dict(type='NumClassCheckHook')]
+
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+
+# disable opencv multithreading to avoid system being overloaded
+opencv_num_threads = 0
+# set multi-process start method as `fork` to speed up the training
+mp_start_method = 'fork'
+
+# Default setting for scaling LR automatically
+#   - `enable` means enable scaling LR automatically
+#       or not by default.
+#   - `base_batch_size` = (8 GPUs) x (2 samples per GPU).
+auto_scale_lr = dict(enable=False, base_batch_size=16)