File size: 3,361 Bytes
feb2918 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 |
import math
import numpy as np
import torch
import torch.nn as nn
from torch.nn import functional as F
from torchvision.ops import roi_align, roi_pool
class VQAHead(nn.Module):
"""MLP Regression Head for VQA.
Args:
in_channels: input channels for MLP
hidden_channels: hidden channels for MLP
dropout_ratio: the dropout ratio for features before the MLP (default 0.5)
pre_pool: whether pre-pool the features or not (True for Aesthetic Attributes, False for Technical Attributes)
"""
def __init__(
self, in_channels=768, hidden_channels=64, dropout_ratio=0.5, pre_pool=False, **kwargs
):
super().__init__()
self.dropout_ratio = dropout_ratio
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.pre_pool = pre_pool
if self.dropout_ratio != 0:
self.dropout = nn.Dropout(p=self.dropout_ratio)
else:
self.dropout = None
self.fc_hid = nn.Conv3d(self.in_channels, self.hidden_channels, (1, 1, 1))
self.fc_last = nn.Conv3d(self.hidden_channels, 1, (1, 1, 1))
self.gelu = nn.GELU()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
def forward(self, x, rois=None):
if self.pre_pool:
x = self.avg_pool(x)
x = self.dropout(x)
qlt_score = self.fc_last(self.dropout(self.gelu(self.fc_hid(x))))
return qlt_score
class VARHead(nn.Module):
"""MLP Regression Head for Video Action Recognition.
Args:
in_channels: input channels for MLP
hidden_channels: hidden channels for MLP
dropout_ratio: the dropout ratio for features before the MLP (default 0.5)
"""
def __init__(self, in_channels=768, out_channels=400, dropout_ratio=0.5, **kwargs):
super().__init__()
self.dropout_ratio = dropout_ratio
self.in_channels = in_channels
self.out_channels = out_channels
if self.dropout_ratio != 0:
self.dropout = nn.Dropout(p=self.dropout_ratio)
else:
self.dropout = None
self.fc = nn.Conv3d(self.in_channels, self.out_channels, (1, 1, 1))
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
def forward(self, x, rois=None):
x = self.dropout(x)
x = self.avg_pool(x)
out = self.fc(x)
return out
class IQAHead(nn.Module):
"""MLP Regression Head for IQA.
Args:
in_channels: input channels for MLP
hidden_channels: hidden channels for MLP
dropout_ratio: the dropout ratio for features before the MLP (default 0.5)
"""
def __init__(
self, in_channels=768, hidden_channels=64, dropout_ratio=0.5, **kwargs
):
super().__init__()
self.dropout_ratio = dropout_ratio
self.in_channels = in_channels
self.hidden_channels = hidden_channels
if self.dropout_ratio != 0:
self.dropout = nn.Dropout(p=self.dropout_ratio)
else:
self.dropout = None
self.fc_hid = nn.Linear(self.in_channels, self.hidden_channels)
self.fc_last = nn.Linear(self.hidden_channels, 1)
self.gelu = nn.GELU()
def forward(self, x):
x = self.dropout(x)
qlt_score = self.fc_last(self.dropout(self.gelu(self.fc_hid(x))))
return qlt_score
|