RWM-dataset

Real-world Model Dataset

After deduplicating the 960 extracted models, there are 245 on-device models left.

There are three sqlite3 database files. In each .db file there are two tables, ai_apk and ai_model. Their schemata are shown as following:

TABLE ai_model:

(hash text primary key, ## model hash

name text NOT NULL, ## model name

ori_apk_hash text NOT NULL, ## model apk hash

model_path text NOT NULL, ## model path (anonymized)

model_type text NOT NULL, ## model type A,B,C

quant_level text NOT NULL, ## quant_level

framework text NOT NULL, ## DL framework

filesize text NOT NULL, ## file size

task text NOT NULL, ## model task

backbone text NOT NULL, ## backbone

input_layer text,

input_size text,

input_type text,

preprocess_param text,

output_layer text,

output_size text,

output_label text,

attack_result text NOT NULL

);

TABLE ai_apk:

(

"hash" text,

"dec_path" text NOT NULL,

"framework" text NOT NULL,

"format_res" text, ## file features

"magic_res" text ## code features

);

Please send us an email to dengzizhuang[at]iie.ac.cn to get the dataset.

======= Instructions about Request Email =======

For Academic Researchers:

If you are a student(or postdoc), please have your advisor (or host) send us an access email. If you are a faculty member, please send us an email from your university email account.

In your email, please include your name, affiliation, and home page (if we do not recognize each other). This information is required only for verification purposes. Please note that if we are unable to determine your identity or affiliation, your request may be ignored.

If your papers or articles use our dataset or our tool, please cite our CCS 2022 paper.

For Industry Researchers:

If you are currently in industry (including research labs), please send us an email from your company’s email account.

In the email, please briefly introduce yourself (e.g., name and title - in case we don’t know each other) and your company.

> Acknowledgement: The application format above is referenced from https://seclab-fudan.github.io/VulScope/