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

Simple Pixelbot for Diablo 2 Resurrected written in python and opencv.

Simple Pixelbot for Diablo 2 Resurrected written in python and opencv. Obviously only use it in offline mode as it is against the TOS of Blizzard to use it in online mode!

468 Jan 03, 2023
SASM - simple crossplatform IDE for NASM, MASM, GAS and FASM assembly languages

SASM (SimpleASM) - простая кроссплатформенная среда разработки для языков ассемблера NASM, MASM, GAS, FASM с подсветкой синтаксиса и отладчиком. В SA

Dmitriy Manushin 5.6k Jan 06, 2023
A PyTorch implementation of the Transformer model in "Attention is All You Need".

Attention is all you need: A Pytorch Implementation This is a PyTorch implementation of the Transformer model in "Attention is All You Need" (Ashish V

Yu-Hsiang Huang 7.1k Jan 04, 2023
Half Instance Normalization Network for Image Restoration

HINet Half Instance Normalization Network for Image Restoration, based on https://github.com/megvii-model/HINet. Dependencies NumPy PyTorch, preferabl

Holy Wu 4 Jun 06, 2022
Match SafeGraph POIs with Data collected through a cultural resource survey in Washington DC.

Match SafeGraph POI data with Cultural Resource Places in Washington DC Match SafeGraph POIs with Data collected through a cultural resource survey in

Changjie Chen 1 Jan 05, 2022
Codes for the compilation and visualization examples to the HIF vegetation dataset

High-impedance vegetation fault dataset This repository contains the codes that compile the "Vegetation Conduction Ignition Test Report" data, which a

1 Dec 12, 2021
Elucidating Robust Learning with Uncertainty-Aware Corruption Pattern Estimation

Elucidating Robust Learning with Uncertainty-Aware Corruption Pattern Estimation Introduction 📋 Official implementation of Explainable Robust Learnin

JeongEun Park 6 Apr 19, 2022
Implementation of SETR model, Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.

SETR - Pytorch Since the original paper (Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.) has no official

zhaohu xing 112 Dec 16, 2022
PantheonRL is a package for training and testing multi-agent reinforcement learning environments.

PantheonRL is a package for training and testing multi-agent reinforcement learning environments. PantheonRL supports cross-play, fine-tuning, ad-hoc coordination, and more.

Stanford Intelligent and Interactive Autonomous Systems Group 57 Dec 28, 2022
I tried to apply the CAM algorithm to YOLOv4 and it worked.

YOLOV4:You Only Look Once目标检测模型在pytorch当中的实现 2021年2月7日更新: 加入letterbox_image的选项,关闭letterbox_image后网络的map得到大幅度提升。 目录 性能情况 Performance 实现的内容 Achievement

55 Dec 05, 2022
Beyond Image to Depth: Improving Depth Prediction using Echoes (CVPR 2021)

Beyond Image to Depth: Improving Depth Prediction using Echoes (CVPR 2021) Kranti Kumar Parida, Siddharth Srivastava, Gaurav Sharma. We address the pr

Kranti Kumar Parida 33 Jun 27, 2022
A toy project using OpenCV and PyMunk

A toy project using OpenCV, PyMunk and Mediapipe the source code for my LindkedIn post It's just a toy project and I didn't write a documentation yet,

Amirabbas Asadi 82 Oct 28, 2022
OCR-D wrapper for detectron2 based segmentation models

ocrd_detectron2 OCR-D wrapper for detectron2 based segmentation models Introduction Installation Usage OCR-D processor interface ocrd-detectron2-segm

Robert Sachunsky 13 Dec 06, 2022
Based on Yolo's low-power, ultra-lightweight universal target detection algorithm, the parameter is only 250k, and the speed of the smart phone mobile terminal can reach ~300fps+

Based on Yolo's low-power, ultra-lightweight universal target detection algorithm, the parameter is only 250k, and the speed of the smart phone mobile terminal can reach ~300fps+

567 Dec 26, 2022
A vision library for performing sliced inference on large images/small objects

SAHI: Slicing Aided Hyper Inference A vision library for performing sliced inference on large images/small objects Overview Object detection and insta

Open Business Software Solutions 2.3k Jan 04, 2023
MAg: a simple learning-based patient-level aggregation method for detecting microsatellite instability from whole-slide images

MAg Paper Abstract File structure Dataset prepare Data description How to use MAg? Why not try the MAg_lib! Trained models Experiment and results Some

Calvin Pang 3 Apr 08, 2022
(CVPR 2022) Pytorch implementation of "Self-supervised transformers for unsupervised object discovery using normalized cut"

(CVPR 2022) TokenCut Pytorch implementation of Tokencut: Self-supervised Transformers for Unsupervised Object Discovery using Normalized Cut Yangtao W

YANGTAO WANG 200 Jan 02, 2023
novel deep learning research works with PaddlePaddle

Research 发布基于飞桨的前沿研究工作,包括CV、NLP、KG、STDM等领域的顶会论文和比赛冠军模型。 目录 计算机视觉(Computer Vision) 自然语言处理(Natrual Language Processing) 知识图谱(Knowledge Graph) 时空数据挖掘(Spa

1.5k Dec 29, 2022
SLAMP: Stochastic Latent Appearance and Motion Prediction

SLAMP: Stochastic Latent Appearance and Motion Prediction Official implementation of the paper SLAMP: Stochastic Latent Appearance and Motion Predicti

Kaan Akan 34 Dec 08, 2022
Graph Neural Networks with Keras and Tensorflow 2.

Welcome to Spektral Spektral is a Python library for graph deep learning, based on the Keras API and TensorFlow 2. The main goal of this project is to

Daniele Grattarola 2.2k Jan 08, 2023