Morris Simons Ai captcha solving model

Labeling the data.

brute forceing for auto labeling the dataset

• how many records in log: 1247
• this is avg time: 26.59 sec
• the amount of max combinations when brutforceing: 36^5 = 60,466,176

Image name cap_1a24d34f483f556e09a850faf9c36932d5bb6ce5 consist of cap_[SHA1_hashvalue]

How to label data

• open config.py

  • if you want to send answers to discord server put in webhook
  • UNLABLED_DIR from where to get data
  • LOGTIME if you want to log how long it takes to bruteforce
  • DATA_DIR under #Universal config to where files will be sent

• open get_train_data.py

  • if no discord comment out send_discord_img(f"{unhashed_guess}_{hash_value}.png", unhashed_guess, operation_time=end) LINE 40
  • to edit combination and len on guess line 30 for guess in itertools.product(string.ascii_lowercase + string.digits, repeat=5):
  • to edit name on file line 28 hash_value = file.split(".png")[0][4:] look up for example on filename

Run relabel_all_captchas to get only answer and remove the old hash value

• files names need to be only the answer my case, fempv.png

Traning data

• open config

  • batchsize for how many times to split data
  • image width for image width on picture
  • image hight for image hight on picture
  • num of workers on your gpu
  • Epochs for how long to run program 50, 100 will have saves
  • cuda and nividia gpu needed for this ai
  • show test samples, "" for all
  • DATA_DIR for where to get data.
  • Modelversion = version name on your model "model_{modelversion}_{epoch}.pt"

• open train

  • line 49 is sensetive for length of filename
  • line 108 to change learning rate
  • e-start on line 98 will get epoch of model (model_"verson"_"50")

Deploy ai

• open config
  • Deployment data for where to get the captchas you want to solve.
  • edit classes.txt to put your all your guesses your program can make.

What all files do summary

• config = settings
• classes = all guesses
• dataset = for image handeling
• deploy ai = for production
• engine = for loss during traning
• get_train_data = for labeling
• log = loggning hash-cracking time
• model_1_50.pt = my model
• model = our ai model
• relabel all captchas = to remove hashvalue and move files
• train - to train ai
• view_data for playing with train model.