Exploring Temporal Coherence for More General Video Face Forgery Detection(FTCN)

Yinglin Zheng, Jianmin Bao, Dong Chen, Ming Zeng, Fang Wen

Accepted by ICCV 2021

Paper

Abstract

Although current face manipulation techniques achieve impressive performance regarding quality and controllability, they are struggling to generate temporal coherent face videos. In this work, we explore to take full advantage of the temporal coherence for video face forgery detection. To achieve this, we propose a novel end-to-end framework, which consists of two major stages. The first stage is a fully temporal convolution network (FTCN). The key insight of FTCN is to reduce the spatial convolution kernel size to 1, while maintaining the temporal convolution kernel size unchanged. We surprisingly find this special design can benefit the model for extracting the temporal features as well as improve the generalization capability. The second stage is a Temporal Transformer network, which aims to explore the long-term temporal coherence. The proposed framework is general and flexible, which can be directly trained from scratch without any pre-training models or external datasets. Extensive experiments show that our framework outperforms existing methods and remains effective when applied to detect new sorts of face forgery videos.

Setup

First setup python environment with pytorch 1.4.0 installed, it's highly recommended to use docker image pytorch/pytorch:1.4-cuda10.1-cudnn7-devel, as the pretrained model and the code might be incompatible with higher version pytorch.

then install dependencies for the experiment:

pip install -r requirements.txt

Test

Inference Using Pretrained Model on Raw Video

Download FTCN+TT model trained on FF++ from here and place it under ./checkpoints folder

python test_on_raw_video.py examples/shining.mp4 output

the output will be a video under folder output named shining.avi

TODO

Release inference code.
Release training code.
Code cleaning.

Acknowledgments

This code borrows heavily from SlowFast.

The face detection network comes from biubug6/Pytorch_Retinaface.

The face alignment network comes from cunjian/pytorch_face_landmark.

Citation

If you use this code for your research, please cite our paper.

@article{zheng2021exploring,
  title={Exploring Temporal Coherence for More General Video Face Forgery Detection},
  author={Zheng, Yinglin and Bao, Jianmin and Chen, Dong and Zeng, Ming and Wen, Fang},
  journal={arXiv preprint arXiv:2108.06693},
  year={2021}
}

按文章中的结构，每个ResBlock中a、b、c三个kernel的size分别应为[1,1,1]，[3,1,1]与[1,1,1]。但代码所输出结构与文中结构不符（如下），或许是理解错误，烦请解惑： res2:

  (s2): ResStage(
    (pathway0_res0): ResBlock(
      (branch1): Conv3d(64, 256, kernel_size=(1, 1, 1), stride=[1, 1, 1], bias=False)
      (branch1_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (branch2): BottleneckTransform(
        (a): Conv3d(64, 64, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(64, 64, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(64, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res1): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(256, 64, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(64, 64, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(64, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res2): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(256, 64, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(64, 64, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(64, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )

res3:

(s3): ResStage(
    (pathway0_res0): ResBlock(
      (branch1): Conv3d(256, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
      (branch1_bn): Sequential(
        (0): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
      )
      (branch2): BottleneckTransform(
        (a): Conv3d(256, 128, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(128, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): Sequential(
          (0): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
        )
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(128, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res1): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(512, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(128, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(128, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res2): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(512, 128, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(128, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(128, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res3): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(512, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(128, 128, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(128, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )

res4:

(s4): ResStage(
    (pathway0_res0): ResBlock(
      (branch1): Conv3d(512, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
      (branch1_bn): Sequential(
        (0): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
      )
      (branch2): BottleneckTransform(
        (a): Conv3d(512, 256, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): Sequential(
          (0): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
        )
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res1): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res2): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 256, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res3): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res4): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 256, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res5): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(256, 256, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(256, 1024, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )

res5:

(s5): ResStage(
    (pathway0_res0): ResBlock(
      (branch1): Conv3d(1024, 2048, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
      (branch1_bn): Sequential(
        (0): BatchNorm3d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
      )
      (branch2): BottleneckTransform(
        (a): Conv3d(1024, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(512, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): Sequential(
          (0): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (1): MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2), padding=0, dilation=1, ceil_mode=False)
        )
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(512, 2048, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res1): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(2048, 512, kernel_size=[3, 1, 1], stride=[1, 1, 1], padding=[1, 0, 0], bias=False)
        (a_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(512, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(512, 2048, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (pathway0_res2): ResBlock(
      (branch2): BottleneckTransform(
        (a): Conv3d(2048, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (a_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (a_relu): ReLU(inplace=True)
        (b): Conv3d(512, 512, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], dilation=[1, 1, 1], bias=False)
        (b_bn): BatchNorm3d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (b_relu): ReLU(inplace=True)
        (c): Conv3d(512, 2048, kernel_size=[1, 1, 1], stride=[1, 1, 1], padding=[0, 0, 0], bias=False)
        (c_bn): BatchNorm3d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )

Code for the paper "Spatio-temporal Self-Supervised Representation Learning for 3D Point Clouds" (ICCV 2021)

Spatio-temporal Self-Supervised Representation Learning for 3D Point Clouds This is the official code implementation for the paper "Spatio-temporal Se

63 Jan 5, 2023

Code for the ICME 2021 paper "Exploring Driving-Aware Salient Object Detection via Knowledge Transfer"

TSOD Code for the ICME 2021 paper "Exploring Driving-Aware Salient Object Detection via Knowledge Transfer" Usage For training, open train_test, run p

2 Dec 23, 2021

VIL-100: A New Dataset and A Baseline Model for Video Instance Lane Detection (ICCV 2021)

Preparation Please see dataset/README.md to get more details about our datasets-VIL100 Please see INSTALL.md to install environment and evaluation too

82 Dec 15, 2022

img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation

img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation Figure 1: We estimate the 6DoF rigid transformation of a 3D face (rendered in si

519 Dec 29, 2022

Code for HLA-Face: Joint High-Low Adaptation for Low Light Face Detection (CVPR21)

HLA-Face: Joint High-Low Adaptation for Low Light Face Detection The official PyTorch implementation for HLA-Face: Joint High-Low Adaptation for Low L

77 Dec 8, 2022

Face Library is an open source package for accurate and real-time face detection and recognition

Face Library Face Library is an open source package for accurate and real-time face detection and recognition. The package is built over OpenCV and us

52 Nov 9, 2022

AI Face Mesh: This is a simple face mesh detection program based on Artificial intelligence.

AI Face Mesh: This is a simple face mesh detection program based on Artificial Intelligence which made with Python. It's able to detect 468 different

1 Jan 13, 2022

Official project website for the CVPR 2021 paper "Exploring intermediate representation for monocular vehicle pose estimation"

EgoNet Official project website for the CVPR 2021 paper "Exploring intermediate representation for monocular vehicle pose estimation". This repo inclu

138 Dec 9, 2022

official Pytorch implementation of ICCV 2021 paper FuseFormer: Fusing Fine-Grained Information in Transformers for Video Inpainting.

FuseFormer: Fusing Fine-Grained Information in Transformers for Video Inpainting By Rui Liu, Hanming Deng, Yangyi Huang, Xiaoyu Shi, Lewei Lu, Wenxiu

77 Dec 27, 2022

Question about the structure of ResNet3D

您好，代码中conv1的kernel size为[5,7,7]，stride为[1,2,2]。而论文中kernel size为[5,1,1]，stride为[1,1,1]。请问，是否可以给出论文中实际使用的，完整的模型结构呢？

temp_kernel[0][0] = [5]
self.s1 = stem_helper.VideoModelStem(
    dim_in=cfg.DATA.INPUT_CHANNEL_NUM,
    dim_out=[width_per_group],
    kernel=[temp_kernel[0][0] + [7, 7]],
    stride=[[1, 2, 2]],
    padding=[[temp_kernel[0][0][0] // 2, 3, 3]],
    norm_module=self.norm_module)

opened by crywang 2

关于模型结构的问题

opened by crywang 1

Bug in test_on_raw_video

            l_post = len(post_module)
            post_module = post_module * (pad_length // l_post + 1)
            post_module = post_module[:pad_length]
            assert len(post_module) == pad_length

            pre_module = inner_index + inner_index[1:-1][::-1]
            l_pre = len(post_module)
            pre_module = pre_module * (pad_length // l_pre + 1)
            pre_module = pre_module[-pad_length:]
            assert len(pre_module) == pad_length

the code

 l_pre = len(post_module)

should be replaced by

 l_pre = len(pre_module)

is it right?

opened by LOOKCC 0

[Official] Exploring Temporal Coherence for More General Video Face Forgery Detection(ICCV 2021)

Related tags

Overview

Exploring Temporal Coherence for More General Video Face Forgery Detection(FTCN)

Paper

Abstract

Setup

Test

Inference Using Pretrained Model on Raw Video

TODO

Acknowledgments

Citation

You might also like...

Code for the paper "Spatio-temporal Self-Supervised Representation Learning for 3D Point Clouds" (ICCV 2021)

Code for the ICME 2021 paper "Exploring Driving-Aware Salient Object Detection via Knowledge Transfer"

VIL-100: A New Dataset and A Baseline Model for Video Instance Lane Detection (ICCV 2021)

img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation

Code for HLA-Face: Joint High-Low Adaptation for Low Light Face Detection (CVPR21)

Face Library is an open source package for accurate and real-time face detection and recognition

AI Face Mesh: This is a simple face mesh detection program based on Artificial intelligence.

Official project website for the CVPR 2021 paper "Exploring intermediate representation for monocular vehicle pose estimation"

official Pytorch implementation of ICCV 2021 paper FuseFormer: Fusing Fine-Grained Information in Transformers for Video Inpainting.

Comments

Question about the structure of ResNet3D

关于模型结构的问题

Bug in test_on_raw_video

Releases(weights)

weights(Sep 1, 2021)

Owner

Learnable Motion Coherence for Correspondence Pruning

Video-face-extractor - Video face extractor with Python

Implementation of temporal pooling methods studied in [ICIP'20] A Comparative Evaluation Of Temporal Pooling Methods For Blind Video Quality Assessment

Cascaded Deep Video Deblurring Using Temporal Sharpness Prior and Non-local Spatial-Temporal Similarity

Change is Everywhere: Single-Temporal Supervised Object Change Detection in Remote Sensing Imagery (ICCV 2021)

Exploring Relational Context for Multi-Task Dense Prediction [ICCV 2021]

A general and strong 3D object detection codebase that supports more methods, datasets and tools (debugging, recording and analysis).

Official Pytorch Implementation of 'Learning Action Completeness from Points for Weakly-supervised Temporal Action Localization' (ICCV-21 Oral)

An official implementation of the paper Exploring Sequence Feature Alignment for Domain Adaptive Detection Transformers

[NeurIPS 2020] Blind Video Temporal Consistency via Deep Video Prior