From 13ab6e980d8487e629582fea8e398b5eab278b37 Mon Sep 17 00:00:00 2001 From: gaotingquan Date: Tue, 27 Jun 2023 11:54:49 +0000 Subject: [PATCH 1/4] unify class_num to num_classes --- passl/models/convnext.py | 4 ++-- passl/models/deit.py | 12 ++++++------ passl/models/dino/dino_vit.py | 8 ++++---- passl/models/dinov2/dinov2_vit.py | 8 ++++---- passl/models/resnet.py | 4 ++-- passl/models/swav.py | 6 +++--- passl/models/vision_transformer.py | 8 ++++---- passl/models/vision_transformer_hybrid.py | 8 ++++---- tasks/ssl/cae/main_finetune.py | 4 ++-- tasks/ssl/mae/main_finetune.py | 4 ++-- 10 files changed, 33 insertions(+), 33 deletions(-) diff --git a/passl/models/convnext.py b/passl/models/convnext.py index cedab3af..723b1375 100644 --- a/passl/models/convnext.py +++ b/passl/models/convnext.py @@ -119,7 +119,7 @@ class ConvNeXt(Model): def __init__( self, in_chans=3, - class_num=1000, + num_classes=1000, depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], drop_path_rate=0., @@ -158,7 +158,7 @@ def __init__( cur += depths[i] self.norm = nn.LayerNorm(dims[-1], epsilon=1e-6) - self.head = nn.Linear(dims[-1], class_num) + self.head = nn.Linear(dims[-1], num_classes) self.apply(self._init_weights) self.head.weight.set_value(self.head.weight * head_init_scale) diff --git a/passl/models/deit.py b/passl/models/deit.py index 0aa71b36..d8ed2ad2 100644 --- a/passl/models/deit.py +++ b/passl/models/deit.py @@ -49,7 +49,7 @@ def __init__(self, img_size=224, patch_size=16, in_chans=3, - class_num=1000, + num_classes=1000, embed_dim=768, depth=12, num_heads=12, @@ -63,7 +63,7 @@ def __init__(self, epsilon=1e-5, **kwargs): super().__init__() - self.class_num = class_num + self.num_classes = num_classes self.num_features = self.embed_dim = embed_dim @@ -101,7 +101,7 @@ def __init__(self, # Classifier head self.head = nn.Linear(embed_dim, - class_num) if class_num > 0 else nn.Identity() + num_classes) if num_classes > 0 else nn.Identity() init.trunc_normal_(self.pos_embed, std=.002) init.trunc_normal_(self.cls_token, std=.002) @@ -197,7 +197,7 @@ class DistilledVisionTransformer(DeitVisionTransformer): def __init__(self, img_size=224, patch_size=16, - class_num=1000, + num_classes=1000, embed_dim=768, depth=12, num_heads=12, @@ -209,7 +209,7 @@ def __init__(self, super().__init__( img_size=img_size, patch_size=patch_size, - class_num=class_num, + num_classes=num_classes, embed_dim=embed_dim, depth=depth, num_heads=num_heads, @@ -229,7 +229,7 @@ def __init__(self, self.head_dist = nn.Linear( self.embed_dim, - self.class_num) if self.class_num > 0 else nn.Identity() + self.num_classes) if self.num_classes > 0 else nn.Identity() init.trunc_normal_(self.dist_token, std=.02) init.trunc_normal_(self.pos_embed, std=.02) diff --git a/passl/models/dino/dino_vit.py b/passl/models/dino/dino_vit.py index dd44c9a9..51470d4a 100644 --- a/passl/models/dino/dino_vit.py +++ b/passl/models/dino/dino_vit.py @@ -280,9 +280,9 @@ def _freeze_norm(self, layer): class LinearClassifier(nn.Layer): """Linear layer to train on top of frozen features""" - def __init__(self, dim, class_num=1000): + def __init__(self, dim, num_classes=1000): super(LinearClassifier, self).__init__() - self.linear = nn.Linear(dim, class_num) + self.linear = nn.Linear(dim, num_classes) normal_(self.linear.weight) zeros_(self.linear.bias) @@ -293,14 +293,14 @@ def forward(self, x): class DINOLinearProbe(DINO): - def __init__(self, class_num=1000, **kwargs): + def __init__(self, num_classes=1000, **kwargs): super().__init__(**kwargs) self.backbone.eval() self.n_last_blocks = self.backbone.n_last_blocks self.avgpool_patchtokens = self.backbone.avgpool_patchtokens embed_dim = self.backbone.embed_dim * (self.n_last_blocks + int(self.avgpool_patchtokens)) - self.linear = LinearClassifier(embed_dim, class_num) + self.linear = LinearClassifier(embed_dim, num_classes) # freeze all layers but the last fc for name, param in self.named_parameters(): diff --git a/passl/models/dinov2/dinov2_vit.py b/passl/models/dinov2/dinov2_vit.py index e4c789fd..c957e3d6 100644 --- a/passl/models/dinov2/dinov2_vit.py +++ b/passl/models/dinov2/dinov2_vit.py @@ -544,9 +544,9 @@ def _freeze_norm(self, layer): class LinearClassifier(nn.Layer): """Linear layer to train on top of frozen features""" - def __init__(self, dim, class_num=1000): + def __init__(self, dim, num_classes=1000): super(LinearClassifier, self).__init__() - self.linear = nn.Linear(dim, class_num) + self.linear = nn.Linear(dim, num_classes) normal_(self.linear.weight) zeros_(self.linear.bias) @@ -557,14 +557,14 @@ def forward(self, x): class DINOv2LinearProbe(DINOv2): - def __init__(self, class_num=1000, **kwargs): + def __init__(self, num_classes=1000, **kwargs): super().__init__(**kwargs) self.backbone.eval() self.n_last_blocks = self.backbone.n_last_blocks self.avgpool_patchtokens = self.backbone.avgpool_patchtokens embed_dim = self.backbone.embed_dim * (self.n_last_blocks + int(self.avgpool_patchtokens)) - self.linear = LinearClassifier(embed_dim, class_num) + self.linear = LinearClassifier(embed_dim, num_classes) # freeze all layers but the last fc for name, param in self.named_parameters(): diff --git a/passl/models/resnet.py b/passl/models/resnet.py index f15f3443..2dead627 100644 --- a/passl/models/resnet.py +++ b/passl/models/resnet.py @@ -55,12 +55,12 @@ def __init__( block, depth=50, width=64, - class_num=1000, + num_classes=1000, with_pool=True, groups=1, zero_init_residual=True, ): - super().__init__(block, depth=depth, width=width, num_classes=class_num, with_pool=with_pool, groups=groups) + super().__init__(block, depth=depth, width=width, num_classes=num_classes, with_pool=with_pool, groups=groups) # Zero-initialize the last BN in each residual branch, # so that the residual branch starts with zeros, and each residual block behaves like an identity. diff --git a/passl/models/swav.py b/passl/models/swav.py index 3ba2dda9..1516a78b 100644 --- a/passl/models/swav.py +++ b/passl/models/swav.py @@ -87,9 +87,9 @@ def _freeze_norm(self, layer): layer._use_global_stats = True class SwAVLinearProbe(SwAV): - def __init__(self, class_num=1000, **kwargs): + def __init__(self, num_classes=1000, **kwargs): super().__init__(**kwargs) - self.linear = RegLogit(class_num) + self.linear = RegLogit(num_classes) self.res_model.eval() # freeze all layers but the last fc @@ -266,7 +266,7 @@ def __init__(self, block, depth, constant_init(sublayer.weight, value=1.0) constant_init(sublayer.bias, value=0.0) - self.encoder = functools.partial(ResNet, block=block, depth=depth)(with_pool=False, class_num=0) + self.encoder = functools.partial(ResNet, block=block, depth=depth)(with_pool=False, num_classes=0) def forward_backbone(self, x): x = self.encoder(x) diff --git a/passl/models/vision_transformer.py b/passl/models/vision_transformer.py index 825e6cc9..534b7b34 100644 --- a/passl/models/vision_transformer.py +++ b/passl/models/vision_transformer.py @@ -257,7 +257,7 @@ def __init__(self, img_size=224, patch_size=16, in_chans=3, - class_num=1000, + num_classes=1000, embed_dim=768, depth=12, num_heads=12, @@ -272,7 +272,7 @@ def __init__(self, representation_size=None, **kwargs): super().__init__() - self.class_num = class_num + self.num_classes = num_classes self.representation_size = representation_size self.num_features = self.embed_dim = embed_dim @@ -322,14 +322,14 @@ def __init__(self, self.tanh = nn.Tanh() self.head = nn.Linear( representation_size, - class_num) if class_num > 0 else nn.Identity() + num_classes) if num_classes > 0 else nn.Identity() init.xavier_uniform_(self.head0.weight) init.zeros_(self.head0.bias) init.xavier_uniform_(self.head.weight) init.constant_(self.head.bias, -10.0) else: self.head = nn.Linear( - embed_dim, class_num) if class_num > 0 else nn.Identity() + embed_dim, num_classes) if num_classes > 0 else nn.Identity() init.zeros_(self.head.weight) init.zeros_(self.head.bias) diff --git a/passl/models/vision_transformer_hybrid.py b/passl/models/vision_transformer_hybrid.py index d066ebb5..d08b772e 100644 --- a/passl/models/vision_transformer_hybrid.py +++ b/passl/models/vision_transformer_hybrid.py @@ -178,7 +178,7 @@ def __init__(self, img_size=224, patch_size=16, in_chans=3, - class_num=1000, + num_classes=1000, embed_dim=768, depth=12, num_heads=12, @@ -193,7 +193,7 @@ def __init__(self, representation_size=None, **kwargs): super().__init__() - self.class_num = class_num + self.num_classes = num_classes self.representation_size = representation_size self.num_features = self.embed_dim = embed_dim @@ -243,14 +243,14 @@ def __init__(self, self.tanh = nn.Tanh() self.head = nn.Linear( representation_size, - class_num) if class_num > 0 else nn.Identity() + num_classes) if num_classes > 0 else nn.Identity() init.xavier_uniform_(self.head0.weight) init.zeros_(self.head0.bias) init.xavier_uniform_(self.head.weight) init.constant_(self.head.bias, -10.0) else: self.head = nn.Linear( - embed_dim, class_num) if class_num > 0 else nn.Identity() + embed_dim, num_classes) if num_classes > 0 else nn.Identity() init.zeros_(self.head.weight) init.zeros_(self.head.bias) diff --git a/tasks/ssl/cae/main_finetune.py b/tasks/ssl/cae/main_finetune.py index f825bd8c..c95da7eb 100644 --- a/tasks/ssl/cae/main_finetune.py +++ b/tasks/ssl/cae/main_finetune.py @@ -455,13 +455,13 @@ def main(args): "alpha": args.mixup, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "class_num": args.nb_classes + "num_classes": args.nb_classes } batch_transform_ops['Cutmix'] = { "alpha": args.cutmix, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "class_num": args.nb_classes + "num_classes": args.nb_classes } mixup_fn = transforms.TransformOpSampler(**batch_transform_ops) diff --git a/tasks/ssl/mae/main_finetune.py b/tasks/ssl/mae/main_finetune.py index 9256a015..76157e1b 100644 --- a/tasks/ssl/mae/main_finetune.py +++ b/tasks/ssl/mae/main_finetune.py @@ -317,13 +317,13 @@ def main(args): "alpha": args.mixup, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "class_num": args.nb_classes + "num_classes": args.nb_classes } batch_transform_ops['Cutmix'] = { "alpha": args.cutmix, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "class_num": args.nb_classes + "num_classes": args.nb_classes } mixup_fn = transforms.TransformOpSampler(**batch_transform_ops) From d0ba7644f589b6471b57100d12a1724a086817f6 Mon Sep 17 00:00:00 2001 From: gaotingquan Date: Wed, 28 Jun 2023 03:36:57 +0000 Subject: [PATCH 2/4] unify class_num to num_classes --- configs/deit/deit-base-p16-pt_in1k-224_2n16c_fp16_o1_dp.yaml | 2 +- configs/lvvit/lvvit_tiny.yaml | 2 +- .../vit-base-p16-pt_in1k-224_4n32c_fp16_o1_dp.yaml | 2 +- .../vit-base-p16-pt_in1k-224_4n32c_fp16_o1_sharding.yaml | 2 +- .../vit-base-p16-pt_in1k-224_4n32c_fp16_o2_dp.yaml | 2 +- .../vit-base-p16-pt_in1k-224_4n32c_fp16_o2_sharding.yaml | 2 +- .../vit-base-p16-pt_in1k-224_4n32c_fp32_dp.yaml | 2 +- configs/vision_transformer/vit-g-p14-pt_in1k-224_1n8c.yaml | 2 +- 8 files changed, 8 insertions(+), 8 deletions(-) diff --git a/configs/deit/deit-base-p16-pt_in1k-224_2n16c_fp16_o1_dp.yaml b/configs/deit/deit-base-p16-pt_in1k-224_2n16c_fp16_o1_dp.yaml index 1135ab80..ee8b6a6f 100644 --- a/configs/deit/deit-base-p16-pt_in1k-224_2n16c_fp16_o1_dp.yaml +++ b/configs/deit/deit-base-p16-pt_in1k-224_2n16c_fp16_o1_dp.yaml @@ -27,7 +27,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.0 drop_path_rate : 0.1 diff --git a/configs/lvvit/lvvit_tiny.yaml b/configs/lvvit/lvvit_tiny.yaml index 7bcfceea..e0a86cd8 100644 --- a/configs/lvvit/lvvit_tiny.yaml +++ b/configs/lvvit/lvvit_tiny.yaml @@ -30,7 +30,7 @@ model: skip_lam: 1 return_dense: True mix_token: True - class_num: 1000 + num_classes: 1000 drop_rate: 0.0 drop_path_rate : 0.1 diff --git a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_dp.yaml b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_dp.yaml index 2583f46f..bc12be5a 100644 --- a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_dp.yaml +++ b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_dp.yaml @@ -27,7 +27,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 diff --git a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_sharding.yaml b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_sharding.yaml index 72ff6433..731be56f 100644 --- a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_sharding.yaml +++ b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o1_sharding.yaml @@ -33,7 +33,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 diff --git a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_dp.yaml b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_dp.yaml index d8066c55..89d2a439 100644 --- a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_dp.yaml +++ b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_dp.yaml @@ -27,7 +27,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 diff --git a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_sharding.yaml b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_sharding.yaml index c905c250..fca4ecd4 100644 --- a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_sharding.yaml +++ b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp16_o2_sharding.yaml @@ -33,7 +33,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 diff --git a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp32_dp.yaml b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp32_dp.yaml index eb19f00d..b75bad95 100644 --- a/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp32_dp.yaml +++ b/configs/vision_transformer/vit-base-p16-pt_in1k-224_4n32c_fp32_dp.yaml @@ -15,7 +15,7 @@ model: mlp_ratio: 4 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 diff --git a/configs/vision_transformer/vit-g-p14-pt_in1k-224_1n8c.yaml b/configs/vision_transformer/vit-g-p14-pt_in1k-224_1n8c.yaml index f41d6bfc..9f91973a 100644 --- a/configs/vision_transformer/vit-g-p14-pt_in1k-224_1n8c.yaml +++ b/configs/vision_transformer/vit-g-p14-pt_in1k-224_1n8c.yaml @@ -15,7 +15,7 @@ model: mlp_ratio: 4.9231 qkv_bias: True epsilon: 1e-6 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 representation_size: 768 label_smoothing: 0.0001 From 69cbcabcfd6479d26a9ee75ac8952583a4976a8d Mon Sep 17 00:00:00 2001 From: gaotingquan Date: Thu, 29 Jun 2023 02:54:33 +0000 Subject: [PATCH 3/4] unify class_num to num_classes in model --- .../convnext/configs/ConvNeXt_base_224_in1k_1n8c_dp_fp32.yaml | 2 +- .../configs/ConvNeXt_base_224_in1k_4n32c_dp_fp16o2.yaml | 2 +- .../convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp32.yaml | 2 +- .../deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml | 2 +- .../deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp32.yaml | 2 +- .../configs/DeiT_base_patch16_224_in1k_2n16c_dp_fp16o2.yaml | 2 +- .../vit/configs/ViT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml | 2 +- .../configs/ViT_base_patch16_384_ft_in1k_1n8c_dp_fp16o2.yaml | 2 +- .../configs/ViT_large_patch16_224_in21k_4n32c_dp_fp16o2.yaml | 2 +- .../configs/ViT_large_patch16_384_in1k_ft_4n32c_dp_fp16o2.yaml | 2 +- .../dino_deit_small_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dino_deit_small_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dino_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dino_vit_base_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dinov2_vit_base_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dinov2_vit_gaint2_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dinov2_vit_large_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../dinov2_vit_small_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../mocov3_deit_base_patch16_224_ft_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../mocov3_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml | 2 +- .../simsiam/configs/simsiam_resnet50_lp_in1k_1n8c_dp_fp32.yaml | 2 +- .../swav/configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml | 2 +- 22 files changed, 22 insertions(+), 22 deletions(-) diff --git a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_1n8c_dp_fp32.yaml b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_1n8c_dp_fp32.yaml index ff684f71..d2e7b382 100644 --- a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_1n8c_dp_fp32.yaml +++ b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_1n8c_dp_fp32.yaml @@ -37,7 +37,7 @@ EMA: Model: name: convnext_base drop_path_rate : 0.5 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp16o2.yaml b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp16o2.yaml index a61ebe22..b37e6c11 100644 --- a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp16o2.yaml +++ b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp16o2.yaml @@ -38,7 +38,7 @@ EMA: Model: name: convnext_base drop_path_rate : 0.5 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp32.yaml b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp32.yaml index 8518d215..f1866215 100644 --- a/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp32.yaml +++ b/tasks/classification/convnext/configs/ConvNeXt_base_224_in1k_4n32c_dp_fp32.yaml @@ -37,7 +37,7 @@ EMA: Model: name: convnext_base drop_path_rate : 0.5 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml index d05badfa..cefb1b19 100644 --- a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml +++ b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml @@ -32,7 +32,7 @@ Model: name: DeiT_base_patch16_224 drop_path_rate : 0.1 drop_rate : 0.0 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp32.yaml b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp32.yaml index 73936477..784ccbdb 100644 --- a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp32.yaml +++ b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_1n8c_dp_fp32.yaml @@ -32,7 +32,7 @@ Model: name: DeiT_base_patch16_224 drop_path_rate : 0.1 drop_rate : 0.0 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_2n16c_dp_fp16o2.yaml b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_2n16c_dp_fp16o2.yaml index 22260732..5a5947a5 100644 --- a/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_2n16c_dp_fp16o2.yaml +++ b/tasks/classification/deit/configs/DeiT_base_patch16_224_in1k_2n16c_dp_fp16o2.yaml @@ -32,7 +32,7 @@ Model: name: DeiT_base_patch16_224 drop_path_rate : 0.1 drop_rate : 0.0 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/classification/vit/configs/ViT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml b/tasks/classification/vit/configs/ViT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml index 54192704..0a97471c 100644 --- a/tasks/classification/vit/configs/ViT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml +++ b/tasks/classification/vit/configs/ViT_base_patch16_224_in1k_1n8c_dp_fp16o2.yaml @@ -30,7 +30,7 @@ DistributedStrategy: # model architecture Model: name: ViT_base_patch16_224 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 # loss function config for traing/eval process diff --git a/tasks/classification/vit/configs/ViT_base_patch16_384_ft_in1k_1n8c_dp_fp16o2.yaml b/tasks/classification/vit/configs/ViT_base_patch16_384_ft_in1k_1n8c_dp_fp16o2.yaml index 0f794cdd..b6fb4c91 100644 --- a/tasks/classification/vit/configs/ViT_base_patch16_384_ft_in1k_1n8c_dp_fp16o2.yaml +++ b/tasks/classification/vit/configs/ViT_base_patch16_384_ft_in1k_1n8c_dp_fp16o2.yaml @@ -31,7 +31,7 @@ DistributedStrategy: # model architecture Model: name: ViT_base_patch16_384 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 # loss function config for traing/eval process diff --git a/tasks/classification/vit/configs/ViT_large_patch16_224_in21k_4n32c_dp_fp16o2.yaml b/tasks/classification/vit/configs/ViT_large_patch16_224_in21k_4n32c_dp_fp16o2.yaml index c58584f3..6d850e55 100644 --- a/tasks/classification/vit/configs/ViT_large_patch16_224_in21k_4n32c_dp_fp16o2.yaml +++ b/tasks/classification/vit/configs/ViT_large_patch16_224_in21k_4n32c_dp_fp16o2.yaml @@ -30,7 +30,7 @@ DistributedStrategy: # model architecture Model: name: ViT_large_patch16_224 - class_num: 21841 + num_classes: 21841 drop_rate: 0.1 # loss function config for traing/eval process diff --git a/tasks/classification/vit/configs/ViT_large_patch16_384_in1k_ft_4n32c_dp_fp16o2.yaml b/tasks/classification/vit/configs/ViT_large_patch16_384_in1k_ft_4n32c_dp_fp16o2.yaml index 033e951b..98796030 100644 --- a/tasks/classification/vit/configs/ViT_large_patch16_384_in1k_ft_4n32c_dp_fp16o2.yaml +++ b/tasks/classification/vit/configs/ViT_large_patch16_384_in1k_ft_4n32c_dp_fp16o2.yaml @@ -31,7 +31,7 @@ DistributedStrategy: # model architecture Model: name: ViT_large_patch16_384 - class_num: 1000 + num_classes: 1000 drop_rate: 0.1 # loss function config for traing/eval process diff --git a/tasks/ssl/dino/configs/dino_deit_small_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dino/configs/dino_deit_small_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml index e20b4c3b..1b61cd5b 100644 --- a/tasks/ssl/dino/configs/dino_deit_small_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dino/configs/dino_deit_small_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 4 avgpool_patchtokens: False - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dino/configs/dino_deit_small_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dino/configs/dino_deit_small_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml index 269ac613..47980857 100644 --- a/tasks/ssl/dino/configs/dino_deit_small_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dino/configs/dino_deit_small_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 4 avgpool_patchtokens: False - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dino/configs/dino_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dino/configs/dino_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml index f32b88e4..e856890c 100644 --- a/tasks/ssl/dino/configs/dino_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dino/configs/dino_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 1 # 0 in small avgpool_patchtokens: True # False in small - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dino/configs/dino_vit_base_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dino/configs/dino_vit_base_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml index d088c162..d0ad569d 100644 --- a/tasks/ssl/dino/configs/dino_vit_base_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dino/configs/dino_vit_base_patch8_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 1 # 0 in small avgpool_patchtokens: True # False in small - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dinov2/configs/dinov2_vit_base_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dinov2/configs/dinov2_vit_base_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml index 2db02133..d6bb8f67 100644 --- a/tasks/ssl/dinov2/configs/dinov2_vit_base_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dinov2/configs/dinov2_vit_base_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 1 avgpool_patchtokens: True - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dinov2/configs/dinov2_vit_gaint2_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dinov2/configs/dinov2_vit_gaint2_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml index fdf7d8df..9f247b24 100644 --- a/tasks/ssl/dinov2/configs/dinov2_vit_gaint2_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dinov2/configs/dinov2_vit_gaint2_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -35,7 +35,7 @@ Model: ffn_layer: swiglufused # n_last_blocks: 1 avgpool_patchtokens: True - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dinov2/configs/dinov2_vit_large_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dinov2/configs/dinov2_vit_large_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml index 637c5f28..10d51839 100644 --- a/tasks/ssl/dinov2/configs/dinov2_vit_large_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dinov2/configs/dinov2_vit_large_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 1 avgpool_patchtokens: True - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/dinov2/configs/dinov2_vit_small_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/dinov2/configs/dinov2_vit_small_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml index 2945d057..c3314534 100644 --- a/tasks/ssl/dinov2/configs/dinov2_vit_small_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/dinov2/configs/dinov2_vit_small_patch14_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -34,7 +34,7 @@ Model: output_dim: 0 n_last_blocks: 1 avgpool_patchtokens: True - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/mocov3/configs/mocov3_deit_base_patch16_224_ft_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/mocov3/configs/mocov3_deit_base_patch16_224_ft_in1k_1n8c_dp_fp16o1.yaml index d70c6647..347eee7f 100644 --- a/tasks/ssl/mocov3/configs/mocov3_deit_base_patch16_224_ft_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/mocov3/configs/mocov3_deit_base_patch16_224_ft_in1k_1n8c_dp_fp16o1.yaml @@ -33,7 +33,7 @@ Model: name: DeiT_base_patch16_224 drop_path_rate : 0.1 drop_rate : 0.0 - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/mocov3/configs/mocov3_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml b/tasks/ssl/mocov3/configs/mocov3_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml index 3903ac48..b720ba5c 100644 --- a/tasks/ssl/mocov3/configs/mocov3_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml +++ b/tasks/ssl/mocov3/configs/mocov3_vit_base_patch16_224_lp_in1k_1n8c_dp_fp16o1.yaml @@ -28,7 +28,7 @@ DistributedStrategy: # model architecture Model: name: mocov3_vit_base_linearprobe - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/simsiam/configs/simsiam_resnet50_lp_in1k_1n8c_dp_fp32.yaml b/tasks/ssl/simsiam/configs/simsiam_resnet50_lp_in1k_1n8c_dp_fp32.yaml index 4ea73b0f..1ee18f54 100644 --- a/tasks/ssl/simsiam/configs/simsiam_resnet50_lp_in1k_1n8c_dp_fp32.yaml +++ b/tasks/ssl/simsiam/configs/simsiam_resnet50_lp_in1k_1n8c_dp_fp32.yaml @@ -28,7 +28,7 @@ DistributedStrategy: # model architecture Model: name: simsiam_resnet50_linearprobe - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: diff --git a/tasks/ssl/swav/configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml b/tasks/ssl/swav/configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml index 59c44277..0045bc2f 100644 --- a/tasks/ssl/swav/configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml +++ b/tasks/ssl/swav/configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml @@ -28,7 +28,7 @@ Model: type: swavresnet50 output_dim: 0 eval_mode: True - class_num: 1000 + num_classes: 1000 # loss function config for traing/eval process Loss: From 20d078671aebafab8a9b59bcba5c67f90a884086 Mon Sep 17 00:00:00 2001 From: gaotingquan Date: Thu, 29 Jun 2023 06:25:38 +0000 Subject: [PATCH 4/4] debug --- tasks/ssl/cae/main_finetune.py | 4 ++-- tasks/ssl/mae/main_finetune.py | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/tasks/ssl/cae/main_finetune.py b/tasks/ssl/cae/main_finetune.py index c95da7eb..f825bd8c 100644 --- a/tasks/ssl/cae/main_finetune.py +++ b/tasks/ssl/cae/main_finetune.py @@ -455,13 +455,13 @@ def main(args): "alpha": args.mixup, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "num_classes": args.nb_classes + "class_num": args.nb_classes } batch_transform_ops['Cutmix'] = { "alpha": args.cutmix, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "num_classes": args.nb_classes + "class_num": args.nb_classes } mixup_fn = transforms.TransformOpSampler(**batch_transform_ops) diff --git a/tasks/ssl/mae/main_finetune.py b/tasks/ssl/mae/main_finetune.py index 76157e1b..9256a015 100644 --- a/tasks/ssl/mae/main_finetune.py +++ b/tasks/ssl/mae/main_finetune.py @@ -317,13 +317,13 @@ def main(args): "alpha": args.mixup, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "num_classes": args.nb_classes + "class_num": args.nb_classes } batch_transform_ops['Cutmix'] = { "alpha": args.cutmix, "prob": args.mixup_switch_prob, "epsilon": args.smoothing, - "num_classes": args.nb_classes + "class_num": args.nb_classes } mixup_fn = transforms.TransformOpSampler(**batch_transform_ops)