Migrations
/
pytorch-image-models
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add_softmax_layer

Browse Source
pull/1229/head
kira7005 3 years ago
parent cbc75c808e
commit 50fd5be983
4 changed files with 2067 additions and 39 deletions
Show Stats Download Patch File Download Diff File

2011
out_log
View File

File diff suppressed because it is too large Load Diff

13
timm/data/textdataset.py
Unescape View File

@ -39,14 +39,19 @@ class TextDataset(Dataset):
All_Videos.sort()
#print(All_Videos)
VideoPath = os.path.join(self.path, All_Videos[idx//32])
#print(VideoPath)
f = open(VideoPath, "r")
feat = idx%32
words = f.read().split()
features = np.float32(words[feat * 4096:feat * 4096 + 4096])
features = torch.tensor(features)
# features = torch.reshape(features, (16, 256))
#print(features.shape)
if len(features) == 0:
print(idx)
print(VideoPath)
features = torch.reshape(features, (16, 256))
# features = torch.reshape(features, (196, 768))
features = torch.reshape(features, (1, 4096))
#features = torch.reshape(features, (1, 4096))
#print(VideoPath)
if VideoPath.find('Normal') == -1:
label = 0
@ -54,9 +59,9 @@ class TextDataset(Dataset):
label = 1
label = torch.tensor(label)
print(features.shape)
#print(features.shape)
#print(features)
print(label.shape)
#print(label.shape)
#print(label)
return features, label

70
timm/models/mlp_mixer.py
Unescape View File

@ -63,7 +63,7 @@ def _cfg(url='', **kwargs):
default_cfgs = dict(
mixer_s32_224=_cfg(),
mixer_s32_224=_cfg(num_classes=2),
mixer_s16_224=_cfg(),
mixer_b32_224=_cfg(),
mixer_b16_224=_cfg(
@ -264,12 +264,13 @@ class MlpMixer(nn.Module):
super().__init__()
self.num_classes = num_classes
self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models
self.initial_fc =nn.Linear(4096, 150528)
self.stem = PatchEmbed(
img_size=img_size, patch_size=patch_size, in_chans=in_chans,
embed_dim=embed_dim, norm_layer=norm_layer if stem_norm else None)
##initial_fc and stem not needed
#self.initial_fc =nn.Linear(4096, 150528)
#self.stem = PatchEmbed(
# img_size=img_size, patch_size=patch_size, in_chans=in_chans,
# embed_dim=embed_dim, norm_layer=norm_layer if stem_norm else None)
# FIXME drop_path (stochastic depth scaling rule or all the same?)
#embed_dim=256
#print("num_classes:",self.num_classes, "embed_dim:", embed_dim)
self.blocks = nn.Sequential(*[
block_layer(
@ -286,23 +287,24 @@ class MlpMixer(nn.Module):
for _ in range(num_blocks)])
"""
self.norm = norm_layer(embed_dim)
# self.head = nn.Linear(embed_dim, self.num_classes) if num_classes > 0 else nn.Identity()
self.head = nn.Sequential(
nn.Linear(embed_dim, self.num_classes),
nn.ReLU(),
nn.Dropout(p=0.3),
nn.Linear(self.num_classes, 1024),
nn.ReLU(),
nn.Dropout(p=0.3),
nn.Linear(1024, 512),
nn.ReLU(),
nn.Dropout(p=0.3),
nn.Linear(512, 256),
nn.ReLU(),
nn.Dropout(p=0.3),
nn.Linear(256, 2)
)
self.sigmoid = nn.Sigmoid()
self.head = nn.Linear(embed_dim, self.num_classes) if num_classes > 0 else nn.Identity()
# self.head = nn.Sequential(
# nn.Linear(embed_dim, self.num_classes),
# nn.ReLU(),
# nn.Dropout(p=0.3),
# nn.Linear(self.num_classes, 1024),
# nn.ReLU(),
# nn.Dropout(p=0.3),
# nn.Linear(1024, 512),
# nn.ReLU(),
# nn.Dropout(p=0.3),
# nn.Linear(512, 256),
# nn.ReLU(),
# nn.Dropout(p=0.3),
# nn.Linear(256, 2)
# )
#self.sigmoid = nn.Sigmoid()
self.sm = nn.Softmax(dim=1)
self.init_weights(nlhb=nlhb)
def init_weights(self, nlhb=False):
@ -318,23 +320,24 @@ class MlpMixer(nn.Module):
def forward_features(self, x):
#x = self.stem(x)
print("In_Model")
#print("In_Model")
x = self.blocks(x)
print(x)
#print(x)
x = self.norm(x)
print(x)
#print(x)
x = x.mean(dim=1)
print(x)
#print(x)
return x
def forward(self, x):
x = self.initial_fc(x)
x = torch.reshape(x, (196, 768))
#x = self.initial_fc(x)
#x = torch.reshape(x, (196, 768))
x = self.forward_features(x)
x = self.head(x)
print(x)
x = self.sigmoid(x)
print(x)
#print(x)
#x = self.sigmoid(x)
#print(x)
x = self.sm(x)
return x
@ -413,7 +416,8 @@ def mixer_s32_224(pretrained=False, **kwargs):
""" Mixer-S/32 224x224
Paper: 'MLP-Mixer: An all-MLP Architecture for Vision' - https://arxiv.org/abs/2105.01601
"""
model_args = dict(patch_size=32, num_blocks=8, embed_dim=512, **kwargs)
#model_args = dict(patch_size=32, num_blocks=8, embed_dim=512, **kwargs)
model_args = dict(patch_size=16, num_blocks=8, embed_dim=256, **kwargs)
model = _create_mixer('mixer_s32_224', pretrained=pretrained, **model_args)
return model

12
train.py
Unescape View File

@ -679,10 +679,11 @@ def train_one_epoch(
data_time_m = AverageMeter()
losses_m = AverageMeter()
print("------Training-------")
model.train()
end = time.time()
print("loader_length=",len(loader))
#print("loader_length=",len(loader))
last_idx = len(loader) - 1
num_updates = epoch * len(loader)
for batch_idx, (input, target) in enumerate(loader):
@ -698,13 +699,15 @@ def train_one_epoch(
with amp_autocast():
#print(model)
print(input.shape)
output = model(input)
print(output.shape)
print(target.shape)
#print(output)
#print(target)
#print(loss_fn)
print(loss_fn)
loss = loss_fn(output, target)
print("loss=", loss)
if not args.distributed:
losses_m.update(loss.item(), input.size(0))
@ -785,6 +788,7 @@ def validate(model, loader, loss_fn, args, amp_autocast=suppress, log_suffix='')
top1_m = AverageMeter()
top5_m = AverageMeter()
print("------Validating-------")
model.eval()
end = time.time()
@ -809,8 +813,12 @@ def validate(model, loader, loss_fn, args, amp_autocast=suppress, log_suffix='')
output = output.unfold(0, reduce_factor, reduce_factor).mean(dim=2)
target = target[0:target.size(0):reduce_factor]
print("output =", output)
print("target=", target)
loss = loss_fn(output, target)
print("eval_loss=", loss)
acc1, acc5 = accuracy(output, target, topk=(1, 5))
print("eval_acc1=", acc1)
if args.distributed:
reduced_loss = reduce_tensor(loss.data, args.world_size)

Write Preview
Loading…
Cancel
Save