[fix]change all bs/lr to 8 gpus setting and fix bugs

minor

configs/recognition/r2plus1d/r2plus1d_r34_3d_32x2x1_180e_kinetics400_rgb.py

@@ -104,16 +104,10 @@ data = dict(
         pipeline=test_pipeline,
         test_mode=True))
 # optimizer
-optimizer = dict(type='SGD', lr=0.15, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='SGD', lr=0.075, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(
-    policy='CosineAnealing',
-    min_lr=0,
-    warmup='linear',
-    warmup_ratio=0.1,
-    warmup_byepoch=True,
-    warmup_iters=40)
+lr_config = dict(policy='CosineAnealing', min_lr=0)
 total_epochs = 180
 checkpoint_config = dict(interval=5)
 evaluation = dict(

configs/recognition/r2plus1d/r2plus1d_r34_8x8x1_180e_kinetics400_rgb.py

@@ -104,16 +104,10 @@ data = dict(
         pipeline=test_pipeline,
         test_mode=True))
 # optimizer
-optimizer = dict(type='SGD', lr=0.4, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='SGD', lr=0.2, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(
-    policy='CosineAnealing',
-    min_lr=0,
-    warmup='linear',
-    warmup_ratio=0.1,
-    warmup_byepoch=True,
-    warmup_iters=40)
+lr_config = dict(policy='CosineAnealing', min_lr=0)
 total_epochs = 180
 checkpoint_config = dict(interval=5)
 evaluation = dict(

configs/recognition/r2plus1d/r2plus1d_r34_video_3d_8x8x1_180e_kinetics400_rgb.py

@@ -88,7 +88,7 @@ test_pipeline = [
     dict(type='ToTensor', keys=['imgs'])
 ]
 data = dict(
-    videos_per_gpu=24,
+    videos_per_gpu=16,
     workers_per_gpu=4,
     train=dict(
         type=dataset_type,
@@ -108,7 +108,7 @@ data = dict(
         pipeline=test_pipeline,
         test_mode=True))
 # optimizer
-optimizer = dict(type='SGD', lr=0.6, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='SGD', lr=0.2, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
 lr_config = dict(

configs/recognition/slowfast/README.md

 # Slowfast
 config | pretrain | top1 acc| top5 acc | gpu_mem(M) | iter time(s) | ckpt | log
 -|-|-|-|-|-|- | -
-slowfast_r50_3d_4x16x1_256e_kinetics400_rgb | None |-|-|9156|1.05|[ckpt]()| [log]()
+slowfast_r50_3d_4x16x1_256e_kinetics400_rgb | None |75.3|92.2|9156|1.05|[ckpt]()| [log]()
 slowfast_r50_3d_8x8x1_256e_kinetics400_rgb|None|-|-|7402|0.84| [ckpt]() | [log]()

configs/recognition/slowfast/slowfast_r50_3d_4x16x1_256e_kinetics400_rgb.py

@@ -15,7 +15,8 @@ model = dict(
             dilations=(1, 1, 1, 1),
             conv1_stride_t=1,
             pool1_stride_t=1,
-            inflate=(0, 0, 1, 1)),
+            inflate=(0, 0, 1, 1),
+            norm_eval=False),
         fast_pathway=dict(
             type='resnet3d',
             depth=50,
@@ -24,7 +25,9 @@ model = dict(
             base_channels=8,
             conv1_kernel=(5, 7, 7),
             conv1_stride_t=1,
-            pool1_stride_t=1)),
+            pool1_stride_t=1,
+            norm_eval=False),
+    ),
     cls_head=dict(
         in_channels=2304,  # 2048+256
         num_classes=400,
@@ -32,7 +35,7 @@ model = dict(
         spatial_type='avg',
         dropout_ratio=0.5))
 train_cfg = None
-test_cfg = dict(average_clips=None)
+test_cfg = dict(average_clips='prob')
 dataset_type = 'RawframeDataset'
 data_root = 'data/kinetics400/rawframes_train/'
 data_root_val = 'data/kinetics400/rawframes_val/'
@@ -111,13 +114,7 @@ data = dict(
 optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(
-    policy='CosineAnealing',
-    min_lr=0,
-    warmup='linear',
-    warmup_ratio=0.01,
-    warmup_byepoch=True,
-    warmup_iters=34)
+lr_config = dict(policy='CosineAnealing', min_lr=0)
 total_epochs = 256
 checkpoint_config = dict(interval=4)
 workflow = [('train', 1)]

configs/recognition/slowfast/slowfast_r50_3d_8x8x1_256e_kinetics400_rgb.py

@@ -15,7 +15,8 @@ model = dict(
             dilations=(1, 1, 1, 1),
             conv1_stride_t=1,
             pool1_stride_t=1,
-            inflate=(0, 0, 1, 1)),
+            inflate=(0, 0, 1, 1),
+            norm_eval=False),
         fast_pathway=dict(
             type='resnet3d',
             depth=50,
@@ -24,7 +25,8 @@ model = dict(
             base_channels=8,
             conv1_kernel=(5, 7, 7),
             conv1_stride_t=1,
-            pool1_stride_t=1)),
+            pool1_stride_t=1,
+            norm_eval=False)),
     cls_head=dict(
         in_channels=2304,  # 2048+256
         num_classes=400,
@@ -32,7 +34,7 @@ model = dict(
         spatial_type='avg',
         dropout_ratio=0.5))
 train_cfg = None
-test_cfg = dict(average_clips=None)
+test_cfg = dict(average_clips='prob')
 dataset_type = 'RawframeDataset'
 data_root = 'data/kinetics400/rawframes_train/'
 data_root_val = 'data/kinetics400/rawframes_val/'
@@ -111,13 +113,7 @@ data = dict(
 optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(
-    policy='CosineAnealing',
-    min_lr=0,
-    warmup='linear',
-    warmup_ratio=0.01,
-    warmup_byepoch=True,
-    warmup_iters=34)
+lr_config = dict(policy='CosineAnealing', min_lr=0)
 total_epochs = 256
 checkpoint_config = dict(interval=4)
 workflow = [('train', 1)]

configs/recognition/slowfast/slowfast_r50_video_3d_4x16x1_256e_kinetics400_rgb.py

@@ -15,7 +15,8 @@ model = dict(
             dilations=(1, 1, 1, 1),
             conv1_stride_t=1,
             pool1_stride_t=1,
-            inflate=(0, 0, 1, 1)),
+            inflate=(0, 0, 1, 1),
+            norm_eval=False),
         fast_pathway=dict(
             type='resnet3d',
             depth=50,
@@ -24,7 +25,8 @@ model = dict(
             base_channels=8,
             conv1_kernel=(5, 7, 7),
             conv1_stride_t=1,
-            pool1_stride_t=1)),
+            pool1_stride_t=1,
+            norm_eval=False)),
     cls_head=dict(
         in_channels=2304,  # 2048+256
         num_classes=400,
@@ -114,13 +116,7 @@ data = dict(
 optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(
-    policy='CosineAnealing',
-    min_lr=0,
-    warmup='linear',
-    warmup_ratio=0.01,
-    warmup_byepoch=True,
-    warmup_iters=34)
+lr_config = dict(policy='CosineAnealing', min_lr=0)
 total_epochs = 256
 checkpoint_config = dict(interval=4)
 workflow = [('train', 1)]

configs/recognition/tsn/tsn_r101_1x1x5_60e_mmit_rgb.py → configs/recognition/tsn/tsn_r101_1x1x5_50e_mmit_rgb.py

@@ -112,14 +112,14 @@ optimizer = dict(
     type='SGD',
     constructor='TSMOptimizerConstructor',
     paramwise_cfg=dict(fc_lr5=True),
-    lr=0.02,
+    lr=0.01,
     momentum=0.9,
     weight_decay=0.0001,
 )
 optimizer_config = dict(grad_clip=dict(max_norm=20, norm_type=2))
 # learning policy
 lr_config = dict(policy='step', step=[20, 40])
-total_epochs = 60
+total_epochs = 50
 checkpoint_config = dict(interval=5)
 evaluation = dict(interval=5, metrics=['mean_average_precision'])
 # yapf:disable
@@ -132,7 +132,7 @@ log_config = dict(
 # runtime settings
 dist_params = dict(backend='nccl')
 log_level = 'INFO'
-work_dir = './work_dirs/tsn_r101_1x1x5_60e_mmit_rgb/'
+work_dir = './work_dirs/tsn_r101_1x1x5_50e_mmit_rgb/'
 load_from = None
 resume_from = None
 workflow = [('train', 1)]

configs/recognition/tsn/tsn_r50_1x1x6_100e_mit_rgb.py

@@ -48,7 +48,12 @@ train_pipeline = [
     dict(type='ToTensor', keys=['imgs', 'label'])
 ]
 val_pipeline = [
-    dict(type='SampleFrames', clip_len=1, frame_interval=1, num_clips=6),
+    dict(
+        type='SampleFrames',
+        clip_len=1,
+        frame_interval=1,
+        num_clips=6,
+        test_mode=True),
     dict(type='FrameSelector', io_backend='memcached', **mc_cfg),
     dict(type='Resize', scale=(-1, 256)),
     dict(type='CenterCrop', crop_size=224),
@@ -59,7 +64,12 @@ val_pipeline = [
     dict(type='ToTensor', keys=['imgs'])
 ]
 test_pipeline = [
-    dict(type='SampleFrames', clip_len=1, frame_interval=1, num_clips=6),
+    dict(
+        type='SampleFrames',
+        clip_len=1,
+        frame_interval=1,
+        num_clips=6,
+        test_mode=True),
     dict(type='FrameSelector', io_backend='memcached', **mc_cfg),
     dict(type='Resize', scale=(-1, 256)),
     dict(type='ThreeCrop', crop_size=256),
@@ -70,8 +80,8 @@ test_pipeline = [
     dict(type='ToTensor', keys=['imgs'])
 ]
 data = dict(
-    videos_per_gpu=16,
-    workers_per_gpu=2,
+    videos_per_gpu=32,
+    workers_per_gpu=4,
     train=dict(
         type=dataset_type,
         ann_file=ann_file_train,
@@ -106,3 +116,4 @@ log_level = 'INFO'
 work_dir = './work_dirs/tsn_r50_1x1x6_100e_mit_rgb'
 load_from = None
 resume_from = None
+workflow = [('train', 1)]

configs/recognition/tsn/tsn_r50_1x1x8_50e_sthv2_rgb.py

@@ -67,7 +67,7 @@ test_pipeline = [
     dict(type='ToTensor', keys=['imgs'])
 ]
 data = dict(
-    videos_per_gpu=4,
+    videos_per_gpu=16,
     workers_per_gpu=2,
     train=dict(
         type=dataset_type,
@@ -85,10 +85,10 @@ data = dict(
         data_prefix=data_root_val,
         pipeline=test_pipeline))
 # optimizer
-optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
 # learning policy
-lr_config = dict(policy='step', step=[40, 80])
+lr_config = dict(policy='step', step=[20, 40])
 total_epochs = 50
 checkpoint_config = dict(interval=5)
 evaluation = dict(