File size: 5,008 Bytes
b32e831
4d10ed1
b32e831
 
 
 
 
 
 
4d10ed1
b32e831
 
 
 
 
 
4d10ed1
b32e831
 
 
 
 
 
 
 
4d10ed1
b32e831
 
 
 
4d10ed1
b32e831
 
 
4d10ed1
 
b32e831
 
 
 
 
 
 
 
4d10ed1
b32e831
4d10ed1
 
 
 
b32e831
4d10ed1
b32e831
 
 
4d10ed1
b32e831
 
 
 
 
 
4d10ed1
b32e831
 
 
 
 
 
 
 
 
 
4d10ed1
b32e831
 
4d10ed1
 
b32e831
 
 
4d10ed1
b32e831
 
 
 
 
 
4d10ed1
b32e831
4d10ed1
b32e831
 
4d10ed1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
# -*- coding: utf-8 -*-
# Author: Gaojian Wang@ZJUICSR; TongWu@ZJUICSR
# --------------------------------------------------------
# This source code is licensed under the Attribution-NonCommercial 4.0 International License.
# You can find the license in the LICENSE file in the root directory of this source tree.
# --------------------------------------------------------

import numpy as np
import torch
import torch.nn.functional as F

import util.misc as misc
from util.metrics import *


@torch.no_grad()
def test_two_class(data_loader, model, device):
    criterion = torch.nn.CrossEntropyLoss()

    # switch to evaluation mode
    model.eval()

    frame_labels = np.array([])  # int label
    frame_preds = np.array([])  # pred logit
    frame_y_preds = np.array([])  # pred int
    video_names_list = list()

    for batch in data_loader:
        images = batch[0]  # torch.Size([BS, C, H, W])
        target = batch[1]  # torch.Size([BS])
        video_name = batch[-1]  # list[BS]
        images = images.to(device, non_blocking=True)
        target = target.to(device, non_blocking=True)

        output = model(images).to(device, non_blocking=True)  # modified
        loss = criterion(output, target)

        frame_pred = (F.softmax(output, dim=1)[:, 1].detach().cpu().numpy())
        frame_preds = np.append(frame_preds, frame_pred)
        frame_y_pred = np.argmax(output.detach().cpu().numpy(), axis=1)
        frame_y_preds = np.append(frame_y_preds, frame_y_pred)

        frame_label = (target.detach().cpu().numpy())
        frame_labels = np.append(frame_labels, frame_label)
        video_names_list.extend(list(video_name))

    # video-level metrics:
    frame_labels_list = frame_labels.tolist()
    frame_preds_list = frame_preds.tolist()
    video_label_list, video_pred_list, video_y_pred_list = get_video_level_label_pred(frame_labels_list, video_names_list, frame_preds_list)

    return frame_preds_list, video_pred_list


@torch.no_grad()
def test_multi_class(data_loader, model, device):
    criterion = torch.nn.CrossEntropyLoss()

    # switch to evaluation mode
    model.eval()

    frame_labels = np.array([])  # int label
    frame_preds = np.empty((0, 4))  # pred logit, initialize as 2D array with 4 columns for 4 classes
    frame_y_preds = np.array([])  # pred int
    video_names_list = list()

    for batch in data_loader:
        images = batch[0]  # torch.Size([BS, C, H, W])
        target = batch[1]  # torch.Size([BS])
        video_name = batch[-1]  # list[BS]
        images = images.to(device, non_blocking=True)
        target = target.to(device, non_blocking=True)

        output = model(images).to(device, non_blocking=True)
        loss = criterion(output, target)

        frame_pred = F.softmax(output, dim=1).detach().cpu().numpy()
        frame_preds = np.append(frame_preds, frame_pred, axis=0)
        frame_y_pred = np.argmax(output.detach().cpu().numpy(), axis=1)
        frame_y_preds = np.append(frame_y_preds, frame_y_pred)

        frame_label = target.detach().cpu().numpy()
        frame_labels = np.append(frame_labels, frame_label)
        video_names_list.extend(list(video_name))

    # video-level metrics:
    frame_labels_list = frame_labels.tolist()
    frame_preds_list = frame_preds.tolist()
    video_label_list, video_pred_list, video_y_pred_list = get_video_level_label_pred_multi_class(frame_labels_list, video_names_list, frame_preds_list)

    return frame_preds_list, video_pred_list


# @torch.no_grad()
# def test_multi_class(data_loader, model, device):
#     criterion = torch.nn.CrossEntropyLoss()
#
#     # switch to evaluation mode
#     model.eval()
#
#     frame_labels = np.array([])  # int label
#     frame_preds = np.array([])  # pred logit
#     frame_y_preds = np.array([])  # pred int
#     video_names_list = list()
#
#     for batch in data_loader:
#         images = batch[0]  # torch.Size([BS, C, H, W])
#         target = batch[1]  # torch.Size([BS])
#         video_name = batch[-1]  # list[BS]
#         images = images.to(device, non_blocking=True)
#         target = target.to(device, non_blocking=True)
#
#         output = model(images).to(device, non_blocking=True)
#         loss = criterion(output, target)
#
#         frame_pred = F.softmax(output, dim=1).detach().cpu().numpy()
#         frame_preds = np.append(frame_preds, frame_pred, axis=0)
#         frame_y_pred = np.argmax(output.detach().cpu().numpy(), axis=1)
#         frame_y_preds = np.append(frame_y_preds, frame_y_pred)
#
#         frame_label = target.detach().cpu().numpy()
#         frame_labels = np.append(frame_labels, frame_label)
#         video_names_list.extend(list(video_name))
#
#     # video-level metrics:
#     frame_labels_list = frame_labels.tolist()
#     frame_preds_list = frame_preds.tolist()
#     video_label_list, video_pred_list, video_y_pred_list = get_video_level_label_pred_multi_class(frame_labels_list, video_names_list, frame_preds_list)
#
#     return frame_preds_list, video_pred_list