QQ Browser 2021 AI Algorithm Competition Track 1 1st Place Program

Overview

2021_QQ_AIAC_Tack1_1st

QQ浏览器2021AI算法大赛赛道一 第1名 方案

paper :

环境

python==3.7.10
torch==1.7.1
transformers==4.5.1
pretrain 需要显存>=24GB 内存>=100GB

数据下载

(1) 视频数据集
视频数据集在官网下载 https://algo.browser.qq.com/
预期主办方会开源数据集,开源后会将地址补上
下载后放到 ./input/data 文件夹
tag_list 为标签的 top1w,官方 baseline 中提供,放到同一文件夹

(2) 预训练模型
预训练模型使用了 https://huggingface.co/hfl/chinese-roberta-wwm-ext-large
请使用 python3 -u download_pretrain_model.py 下载

步骤代码

(1) 预训练 + finetune
脚本命令:sh train.sh
时间算力:单模在 1 张 a100 上大约需要 pretrain(2 day),finetune(2 hour)
输出文件:每个单模的 checkpoint 保存在 jobN/model_finetune_1.pth
备注:各个单模间没有前后依赖关系,每个任务需要一张单卡,有多卡可以并行训练各个单模

(2) 代码结构说明
download_pretrain_model.py : 下载预训练模型的脚本
ensemble.py : 融合的脚本
job1-job6 : 六个模型训练任务,其文件结构完全一致,各 job 之间主要差别在预训练设置上
注:job1在赛后额外补充了一些代码注释
jobN/pretrain.py 预训练脚本
jobN/finetune.py finetune脚本
jobN/data 数据预处理部分,包含 dataset、mask token 等
jobN/config 包含 pretrain 与 finetune 的一些超参配置
jobN/qqmodel/qq_uni_model.py 模型定义

简介

简要介绍的 ppt 请参考 Introduction.pdf

模型简介

多模态模型结构与参数量和 Bert-large 一致,
layer=24, hidden_size=1024, num_attention_heads=16。
其输入为[CLS] Video_frame [SEP] Video_title [SEP]。
frame_feature 通过 fc 降维为 1024 维,与 text 的 emb 拼接。
Input_emb -> TransformerEncoder * 24 -> Pooling -> Fc -> Video_emb

预训练

预训练采用了 Tag classify, Mask language model, Mask frame model 三个任务

(1) Video tag classify 任务
tag 为人工标注的视频标签,pointwise 和 pairwise 数据集合中提供。
和官方提供的 baseline 一致,我们采用了出现频率前1w 的tag 做多标签分类任务。
Bert 最后一层的 [CLS] -> fc 得到 tag 的预测标签,与真实标签计算 BCE loss

(2) Mask language model 任务
与常见的自然语言处理 mlm 预训练方法相同,对 text 随机 15% 进行 mask,预测 mask 词。
多模态场景下,结合视频的信息预测 mask 词,可以有效融合多模态信息。

(3) Mask frame model 任务
对 frame 的随机 15% 进行 mask,mask 采用了全 0 的向量填充。
考虑到 frame 为连续的向量,难以类似于 mlm 做分类任务。
借鉴了对比学习思路,希望 mask 的预测帧在整个 batch 内的所有帧范围内与被 mask 的帧尽可能相似。
采用了 Nce loss,最大化 mask 帧和预测帧的互信息

(4) 多任务联合训练
预训练任务的 loss 采用了上述三个任务 loss 的加权和,
L = L(tag) * 1250 / 3 + L(mlm) / 3.75 + L(mfm) / 9
tag 梯度量级比较小,因此乘以了较大的权重。
注:各任务合适的权重对下游 finetune 的效果影响比较大。

(5) 预训练 Setting
初始化:bert 初始化权重来自于在中文语料预训练过的开源模型 https://huggingface.co/hfl/chinese-roberta-wwm-ext-large
数据集:预训练使用了 pointwise 和 pairwise 集合,部分融合模型中加上了 test 集合(只有 mlm 和 mfm 任务)
超参:batch_size=128, epoch=40, learning_rate=5e-5, scheduler=warmup_with_cos_decay, warum_ratio=0.06
注:预训练更多的 epoch 对效果提升比较大,从10 epoch 提升至 20 epoch 对下游任务 finetune 效果提升显著。

Finetune

(1) 下游任务
视频 pair 分别通过 model 得到 256维 embedding,两个 embedding 的 cos 相似度与人工标注标签计算 mse

(2) Finetune header
实验中发现相似度任务中,使用 mean_pooling 或者 attention_pooling 聚合最后一层 emb 接 fc 层降维效果较好。

(3) Label normalize
评估指标为 spearman,考查预测值和实际值 rank 之间的相关性,因此对人工标注 label 做了 rank 归一化。
即 target = scipy.stats.rankdata(target, 'average')

(4) Finetune Setting
数据集:训练集使用了 pairwise 中 (id1%5!=0) | (id2%5 !=0) 的部分约 6.5w,验证集使用了(id1%5==0) & (id2%5==0) 的部分约 2.5k
超参:batch_size=32, epoch=10, learning_rate=1e-5, scheduler=warmup_with_cos_decay, warum_ratio=0.06

Ensemble

(1) 融合的方法
采用了 weighted concat -> svd 降维 方法进行融合。实验中发现这种方法降维效果折损较小。
concat_vec = [np.sqrt(w1) * emb1, np.sqrt(w2) * emb2, np.sqrt(w3) * emb3 ...]
svd_vec = SVD(concat_vec, 256)

(2) 融合的模型
最终的提交融合了六个模型。 模型都使用了 bert-large 这种结构,均为迭代过程中产出的模型,各模型之间只有微小的 diff,各个模型加权权重均为 1/6。
下面表格中列出了各模型的diff部分,验证集mse,验证集spearman

jobid ensemble-weight detail val-spearman val-mse
job1 1/6 base 0.886031 0.028813
job2 1/6 预训练tag分类任务为mean_pooling+fc 0.884257 0.029493
job3 1/6 预训练任务无 mfm 0.883843 0.029248
job4 1/6 预训练数据为 (point + pair)shuf-40epoch => pair-5epoch 0.885397 0.029059
job5 1/6 预训练数据为 (point-shuf => pair-shuf => test-shuf)-32epoch 0.885795 0.028866
job6 1/6 预训练 mlm/mfm mask概率调整为25% 0.886289 0.029039

(3) 单模型的效果与融合的效果
单模的测试集成绩约在 0.836
融合两个模型在 0.845
融合三个模型在 0.849
融合五个模型在 0.852

Newt - a Gaussian process library in JAX.

Newt __ \/_ (' \`\ _\, \ \\/ /`\/\ \\ \ \\

AaltoML 0 Nov 02, 2021
A PyTorch Image-Classification With AlexNet And ResNet50.

PyTorch 图像分类 依赖库的下载与安装 在终端中执行 pip install -r -requirements.txt 完成项目依赖库的安装 使用方式 数据集的准备 STL10 数据集 下载:STL-10 Dataset 存储位置:将下载后的数据集中 train_X.bin,train_y.b

FYH 4 Feb 22, 2022
A curated list of the latest breakthroughs in AI (in 2021) by release date with a clear video explanation, link to a more in-depth article, and code.

2021: A Year Full of Amazing AI papers- A Review 📌 A curated list of the latest breakthroughs in AI by release date with a clear video explanation, l

Louis-François Bouchard 2.9k Dec 31, 2022
PyTorch common framework to accelerate network implementation, training and validation

pytorch-framework PyTorch common framework to accelerate network implementation, training and validation. This framework is inspired by works from MML

Dongliang Cao 3 Dec 19, 2022
Source Code for DialogBERT: Discourse-Aware Response Generation via Learning to Recover and Rank Utterances (https://arxiv.org/pdf/2012.01775.pdf)

DialogBERT This is a PyTorch implementation of the DialogBERT model described in DialogBERT: Neural Response Generation via Hierarchical BERT with Dis

Xiaodong Gu 67 Jan 06, 2023
Download and preprocess popular sequential recommendation datasets

Sequential Recommendation Datasets This repository collects some commonly used sequential recommendation datasets in recent research papers and provid

125 Dec 06, 2022
For auto aligning, cropping, and scaling HR and LR images for training image based neural networks

ImgAlign For auto aligning, cropping, and scaling HR and LR images for training image based neural networks Usage Make sure OpenCV is installed, 'pip

15 Dec 04, 2022
IndoNLI: A Natural Language Inference Dataset for Indonesian

IndoNLI: A Natural Language Inference Dataset for Indonesian This is a repository for data and code accompanying our EMNLP 2021 paper "IndoNLI: A Natu

15 Feb 10, 2022
Code and Resources for the Transformer Encoder Reasoning Network (TERN)

Transformer Encoder Reasoning Network Code for the cross-modal visual-linguistic retrieval method from "Transformer Reasoning Network for Image-Text M

Nicola Messina 53 Dec 30, 2022
Efficiently computes derivatives of numpy code.

Note: Autograd is still being maintained but is no longer actively developed. The main developers (Dougal Maclaurin, David Duvenaud, Matt Johnson, and

Formerly: Harvard Intelligent Probabilistic Systems Group -- Now at Princeton 6.1k Jan 08, 2023
Pixray is an image generation system

Pixray is an image generation system

pixray 883 Jan 07, 2023
DL & CV-based indicator toolset for the vehicle drivers via live dash-cam footage.

Vehicle Indicator Toolset Deep Learning and Computer Vision based indicator toolset for vehicle drivers using live dash-cam footages. Tracking of vehi

Alex Xu 12 Dec 28, 2021
✂️ EyeLipCropper is a Python tool to crop eyes and mouth ROIs of the given video.

EyeLipCropper EyeLipCropper is a Python tool to crop eyes and mouth ROIs of the given video. The whole process consists of three parts: frame extracti

Zi-Han Liu 9 Oct 25, 2022
ColossalAI-Examples - Examples of training models with hybrid parallelism using ColossalAI

ColossalAI-Examples This repository contains examples of training models with Co

HPC-AI Tech 185 Jan 09, 2023
FMA: A Dataset For Music Analysis

FMA: A Dataset For Music Analysis Michaël Defferrard, Kirell Benzi, Pierre Vandergheynst, Xavier Bresson. International Society for Music Information

Michaël Defferrard 1.8k Dec 29, 2022
An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge.

Bottom-Up and Top-Down Attention for Visual Question Answering An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge. The

Hengyuan Hu 731 Jan 03, 2023
Cycle Consistent Adversarial Domain Adaptation (CyCADA)

Cycle Consistent Adversarial Domain Adaptation (CyCADA) A pytorch implementation of CyCADA. If you use this code in your research please consider citi

Hyunwoo Ko 2 Jan 10, 2022
PiCIE: Unsupervised Semantic Segmentation using Invariance and Equivariance in clustering (CVPR2021)

PiCIE: Unsupervised Semantic Segmentation using Invariance and Equivariance in Clustering Jang Hyun Cho1, Utkarsh Mall2, Kavita Bala2, Bharath Harihar

Jang Hyun Cho 164 Dec 30, 2022
Code from the paper "High-Performance Brain-to-Text Communication via Handwriting"

High-Performance Brain-to-Text Communication via Handwriting Overview This repo is associated with this manuscript, preprint and dataset. The code can

Francis R. Willett 306 Jan 03, 2023
Unified tracking framework with a single appearance model

Paper: Do different tracking tasks require different appearance model? [ArXiv] (comming soon) [Project Page] (comming soon) UniTrack is a simple and U

ZhongdaoWang 300 Dec 24, 2022