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

Code for the paper "Generative design of breakwaters usign deep convolutional neural network as a surrogate model"

Generative design of breakwaters usign deep convolutional neural network as a surrogate model This repository contains the code for the paper "Generat

2 Apr 10, 2022
SOFT: Softmax-free Transformer with Linear Complexity, NeurIPS 2021 Spotlight

SOFT: Softmax-free Transformer with Linear Complexity SOFT: Softmax-free Transformer with Linear Complexity, Jiachen Lu, Jinghan Yao, Junge Zhang, Xia

Fudan Zhang Vision Group 272 Dec 25, 2022
SSD-based Object Detection in PyTorch

SSD-based Object Detection in PyTorch 서강대학교 현대모비스 SW 프로그램에서 진행한 인공지능 프로젝트입니다. Jetson nano를 이용해 pre-trained network를 fine tuning시켜 차량 및 신호등 인식을 구현하였습니다

Haneul Kim 1 Nov 16, 2021
Generate indoor scenes with Transformers

SceneFormer: Indoor Scene Generation with Transformers Initial code release for the Sceneformer paper, contains models, train and test scripts for the

Chandan Yeshwanth 110 Dec 06, 2022
Task-based end-to-end model learning in stochastic optimization

Task-based End-to-end Model Learning in Stochastic Optimization This repository is by Priya L. Donti, Brandon Amos, and J. Zico Kolter and contains th

CMU Locus Lab 164 Dec 29, 2022
A collection of Google research projects related to Federated Learning and Federated Analytics.

Federated Research Federated Research is a collection of research projects related to Federated Learning and Federated Analytics. Federated learning i

Google Research 483 Jan 05, 2023
A custom DeepStack model for detecting 16 human actions.

DeepStack_ActionNET This repository provides a custom DeepStack model that has been trained and can be used for creating a new object detection API fo

MOSES OLAFENWA 16 Nov 11, 2022
Tracking Progress in Question Answering over Knowledge Graphs

Tracking Progress in Question Answering over Knowledge Graphs Table of contents Question Answering Systems with Descriptions The QA Systems Table cont

Knowledge Graph Question Answering 47 Jan 02, 2023
PolyTrack: Tracking with Bounding Polygons

PolyTrack: Tracking with Bounding Polygons Abstract In this paper, we present a novel method called PolyTrack for fast multi-object tracking and segme

Gaspar Faure 13 Sep 15, 2022
NVTabular is a feature engineering and preprocessing library for tabular data designed to quickly and easily manipulate terabyte scale datasets used to train deep learning based recommender systems.

NVTabular is a feature engineering and preprocessing library for tabular data designed to quickly and easily manipulate terabyte scale datasets used to train deep learning based recommender systems.

880 Jan 07, 2023
A curated list of awesome projects and resources related fastai

A curated list of awesome projects and resources related fastai

Tanishq Abraham 138 Dec 22, 2022
Anti-UAV base on PaddleDetection

Paddle-Anti-UAV Anti-UAV base on PaddleDetection Background UAVs are very popular and we can see them in many public spaces, such as parks and playgro

Qingzhong Wang 2 Apr 20, 2022
Official PyTorch Implementation of SSMix (Findings of ACL 2021)

SSMix: Saliency-based Span Mixup for Text Classification (Findings of ACL 2021) Official PyTorch Implementation of SSMix | Paper Abstract Data augment

Clova AI Research 52 Dec 27, 2022
Digan - Official PyTorch implementation of Generating Videos with Dynamics-aware Implicit Generative Adversarial Networks

DIGAN (ICLR 2022) Official PyTorch implementation of "Generating Videos with Dyn

Sihyun Yu 147 Dec 31, 2022
Official Code Release for "TIP-Adapter: Training-free clIP-Adapter for Better Vision-Language Modeling"

Official Code Release for "TIP-Adapter: Training-free clIP-Adapter for Better Vision-Language Modeling" Pipeline of Tip-Adapter Tip-Adapter can provid

peng gao 187 Dec 28, 2022
ALBERT-pytorch-implementation - ALBERT pytorch implementation

ALBERT-pytorch-implementation developing... 모델의 개념이해를 돕기 위한 구현물로 현재 변수명을 상세히 적었고

BG Kim 3 Oct 06, 2022
A script depending on VASP output for calculating Fermi-Softness.

Fermi softness calculation for Vienna Ab initio Simulation Package (VASP) Update 1.1.0: Big update: Rewrote the code. Use Bader atomic division instea

qslin 11 Nov 08, 2022
[ACM MM 2021] Multiview Detection with Shadow Transformer (and View-Coherent Data Augmentation)

Multiview Detection with Shadow Transformer (and View-Coherent Data Augmentation) [arXiv] [paper] @inproceedings{hou2021multiview, title={Multiview

Yunzhong Hou 27 Dec 13, 2022
Christmas face app for Decathlon xmas coding party!

Christmas Face Application Use this library to create the perfect picture for your christmas cards! Done by Hasib Zunair, Guillaume Brassard and Samue

Hasib Zunair 4 Dec 20, 2021
CONditionals for Ordinal Regression and classification in tensorflow

Condor Ordinal regression in Tensorflow Keras Tensorflow Keras implementation of CONDOR Ordinal Regression (aka ordinal classification) by Garrett Jen

9 Jul 31, 2022