Functions — github.com/CandiceD17/Bottleneck-Transformers-for-Visual-Recognition

hub / github.com/CandiceD17/Bottleneck-Transformers-for-Visual-Recognition / functions

Functions42 in github.com/CandiceD17/Bottleneck-Transformers-for-Visual-Recognition

⨍ Functions 42 ◇ Types & classes 9

↓ 10 callersMethodupdate

(self, val, n=1)

util/utils.py:155

↓ 4 callersMethod_make_layer

(self, block, planes: int, blocks: int, stride: int = 1, dilate: bool = False)

model/resnet_backbone.py:150

↓ 3 callersFunctionconv1x1

1x1 convolution

model/resnet_backbone.py:19

↓ 2 callersMethoddisplay

(self, batch)

util/utils.py:172

↓ 2 callersFunctionrelative_logits_1d

q: [B, Nh, H, W, d] rel_k: [2W - 1, d] Computes relative logits along one dimension. The details of relative position is explained in

model/rel_pos_embedding.py:35

↓ 1 callersMethod__init__

dim: channels in feature map dim_out: output channels for feature map

model/botnet.py:10

↓ 1 callersMethod__init__

( self, inplanes: int, planes: int, stride: int = 1, downsample: Optio

model/resnet_backbone.py:33

↓ 1 callersMethod__init__

(self, height, width, dim_head)

model/rel_pos_embedding.py:53

↓ 1 callersMethod_forward_impl

(self, x: Tensor)

model/resnet_backbone.py:175

↓ 1 callersMethod_get_batch_fmtstr

(self, num_batches)

util/utils.py:177

↓ 1 callersFunctionconv3x3

3x3 convolution with padding

model/resnet_backbone.py:13

↓ 1 callersFunctionexpand_dim

Expand dims for t at dim to k

model/rel_pos_embedding.py:7

↓ 1 callersFunctionrel_to_abs

x: [B, Nh * H, L, 2L - 1] Convert relative position between the key and query to their absolute position respectively. Tensowflow source

model/rel_pos_embedding.py:17

↓ 1 callersMethodreset

(self)

util/utils.py:149

↓ 1 callersFunctiontrain_epoch

(train_loader, net, criterion, optimizer, scheduler, cur_epoch, rank)

train.py:17

↓ 1 callersFunctiontrain_model

()

train.py:103

↓ 1 callersFunctionvalidate

(val_loader, net, criterion, cur_epoch, rank)

train.py:64

Method__init__

dim: number of channels of feature map fmap_size: [H, W] dim_qk: vector dimension for q, k dim_v: vector dimension fo

model/self_attention_block.py:9

Method__init__

dim: channels in feature map fmap_size: [H, W]

model/botnet.py:70

Method__init__

( self, fmap_size, layers: List[int] = [3, 4, 6, 3], botnet: bool = False,

model/resnet_backbone.py:87

Method__init__

(self, height, width, dim_head)

model/rel_pos_embedding.py:68

Method__init__

(self, name, fmt=":f")

util/utils.py:144

Method__init__

(self, num_batches, meters, prefix="")

Computes the accuracy over the k top predictions for the specified values of k

util/utils.py:183

Functionconstruct_loader

Constructs the data loader for the given dataset.

util/utils.py:80

Functionconstruct_optimizer

(model)

util/utils.py:130

Methodforward

featuremap: [B, d_in, H, W] Output: [B, H, W, head * d_v]

model/self_attention_block.py:39

model/resnet_backbone.py:59

Methodforward

(self, x: Tensor)

model/resnet_backbone.py:193

Methodforward

(self, q)

model/rel_pos_embedding.py:60

Methodforward

(self, q)

model/rel_pos_embedding.py:76

Functionget_lr_scheduler

Functionload_cfg_fom_args

Load config from command line arguments and set any specified options.

util/config.py:49

Functionreset_cfg

Reset config to initial state.

util/config.py:44

Functionsave_checkpoint

(state, is_best, filename="ckpt.pth.tar")

util/utils.py:221

Functionscaled_all_reduce

Performs the scaled all_reduce operation on the provided tensors. The input tensors are modified in-place. Currently supports only the sum red

util/utils.py:56

Functionsetup_distributed

Initialize distributed training environment. support both slurm and torch.distributed.launch