Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

update import paths

Browse Source
This commit is contained in:
Vladimir Mandic
2023-02-08 11:53:39 -05:00
parent b31ce33fb6
commit fd668e3941
22 changed files with 227 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cli/modules/bench.py
+2 -7
View File
@@ -6,15 +6,10 @@ import asyncio
import base64
import io
import json
import os
import sys
import time
from PIL import Image
sys.path.append(os.path.join(os.path.dirname(__file__), 'modules'))
import modules.sdapi as sdapi
from modules.util import Map, log
import sdapi as sdapi
from util import Map, log
options = Map({
cli/modules/embedding-preview.py
-1
View File
@@ -11,7 +11,6 @@ import argparse
from pathlib import Path
from PIL import Image
from inspect import getsourcefile
from util import Map, log
from sdapi import getsync, postsync
from grid import grid
cli/extract-lora.py → cli/modules/extract-lora.py
+2 -2
View File
@@ -12,11 +12,11 @@ import argparse
import torch
import transformers
from tqdm import tqdm
from util import log
sys.path.append(os.path.join(os.path.dirname(__file__), '..', 'modules', 'lora'))
sys.path.append(os.path.join(os.path.dirname(__file__), '..', '..', 'modules', 'lora'))
import library.model_util as model_util
import networks.lora as lora
from modules.util import log
def svd(args): # pylint: disable=redefined-outer-name
cli/modules/grid.py
-1
View File
@@ -8,7 +8,6 @@ import argparse
import math
import logging
from pathlib import Path
import filetype
from PIL import Image, ImageDraw, ImageFont
from util import log
cli/modules/image-watermark.py
+1 -1
View File
@@ -9,7 +9,7 @@ from imwatermark import WatermarkEncoder, WatermarkDecoder
from PIL import Image
from PIL.ExifTags import TAGS
from PIL.TiffImagePlugin import ImageFileDirectory_v2
from modules.util import log, Map
from util import log, Map
import piexif
import piexif.helper
cli/modules/interrogate.py
-2
View File
@@ -8,10 +8,8 @@ import base64
import sys
import os
import asyncio
import filetype
from PIL import Image
from util import log, Map
import sdapi as sdapi
cli/modules/models-preview.py
+3 -4
View File
@@ -6,13 +6,12 @@ import time
import asyncio
import argparse
from pathlib import Path
from util import Map, log
from sdapi import get, post, close
from grid import grid
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
sys.path.append(os.path.join(os.path.dirname(__file__), 'modules'))
from generate import sd, generate
from modules.util import Map, log
from modules.sdapi import get, post, close
from modules.grid import grid
default = 'sd-v15-runwayml.ckpt [cc6cb27103]'
cli/modules/process.py
+8 -8
View File
@@ -25,14 +25,12 @@ import base64
import pathlib
import argparse
import logging
import filetype
import numpy as np
import mediapipe as mp
from PIL import Image, ImageOps
from skimage.metrics import structural_similarity as ssim
from scipy.stats import beta
from util import log, Map
from sdapi import postsync
@@ -40,6 +38,7 @@ from sdapi import postsync
params = Map({
'src': '', # source folder
'dst': '', # destination folder
'format': '.png', # image format
'extract_face': True, # extract face from image
'extract_body': True, # extract face from image
'clear_dst': True, # remove all files from destination at the start
@@ -50,15 +49,15 @@ params = Map({
'face_pad': 0.07, # pad face image percentage
'face_model': 1, # which face model to use 0/close-up 1/standard
'face_blur_score': 1.5, # max score for face blur detection
'face_range_score': 0.3, # min score for face dynamic range detection
'face_range_score': 0.2, # min score for face dynamic range detection
'body_score': 0.9, # min body detection score
'body_visibility': 0.5, # min visibility score for each detected body part
'body_parts': 15, # min number of detected body parts with sufficient visibility
'body_pad': 0.2, # pad body image percentage
'body_model': 2, # body model to use 0/low 1/medium 2/high
'body_blur_score': 1.8, # max score for body blur detection
'body_range_score': 0.3, # min score for body dynamic range detection
'segmentation_face': True, # segmentation enabled
'body_range_score': 0.2, # min score for body dynamic range detection
'segmentation_face': False, # segmentation enabled
'segmentation_body': False, # segmentation enabled
'segmentation_model': 0, # segmentation model 0/general 1/landscape
'segmentation_background': (192, 192, 192), # segmentation background color
@@ -263,7 +262,7 @@ def interrogate(img, fn):
res = postsync('/sdapi/v1/interrogate', json)
caption = res.caption if 'caption' in res else ''
log.info({ 'interrogate': caption })
file = fn.replace('.jpg', '.txt')
file = fn.replace(params.format, '.txt')
f = open(file, 'w')
f.write(caption)
f.close()
@@ -278,7 +277,7 @@ def process_file(f: str, dst: str = None, preview: bool = False, offline: bool =
else:
dir = dst
base = os.path.basename(f).split('.')[0]
fn = os.path.join(dir, str(i[what]).rjust(3, '0') + '-' + what + '-' + base + '.jpg')
fn = os.path.join(dir, str(i[what]).rjust(3, '0') + '-' + what + '-' + base + params.format)
# log.debug({ 'save': fn })
if not preview:
img.save(fn)
@@ -318,6 +317,7 @@ def process_file(f: str, dst: str = None, preview: bool = False, offline: bool =
log.debug({ 'no body': f })
image.close()
return i
def process_images(src: str, dst: str, args = None):
params.src = src
@@ -337,7 +337,7 @@ def process_images(src: str, dst: str, args = None):
for root, _sub_dirs, files in os.walk(src):
for f in files:
process_file(os.path.join(root, f), dst)
return i
if __name__ == '__main__':
# log.setLevel(logging.DEBUG)
cli/modules/prompt-ideas.py
-2
View File
@@ -6,9 +6,7 @@ model from: <https://huggingface.co/FredZhang7/distilgpt2-stable-diffusion-v2>
import logging
import argparse
from transformers import GPT2Tokenizer, GPT2LMHeadModel
from util import log
cli/modules/prompt-promptist.py
-2
View File
@@ -5,9 +5,7 @@ use microsoft promptist to beautify prompt
"""
import sys
from transformers import AutoModelForCausalLM, AutoTokenizer
from util import log
cli/modules/sdapi.py
-1
View File
@@ -9,7 +9,6 @@ import asyncio
import logging
import requests
import sys
from util import Map, log
cli/modules/train-lora.py
Executable
+186
View File
@@ -0,0 +1,186 @@
#!/bin/env python
"""
Extract approximating LoRA by SVD from two SD models
Based on: <https://github.com/kohya-ss/sd-scripts/blob/main/networks/train_network.py>
"""
import os
import sys
import argparse
import tempfile
import transformers
from pathlib import Path
from util import log, Map
from process import process_file
sys.path.append(os.path.join(os.path.dirname(__file__), '..', '..', 'modules', 'lora'))
from train_network import train
options = Map({
"v2": False,
"v_parameterization": False,
"pretrained_model_name_or_path": "/mnt/d/Models/stable-diffusion/sd-v15-runwayml.ckpt",
"train_data_dir": "/tmp/rreid/img",
"shuffle_caption": False,
"caption_extension": ".txt",
"caption_extention": None,
"keep_tokens": None,
"color_aug": False,
"flip_aug": False,
"face_crop_aug_range": None,
"random_crop": False,
"debug_dataset": False,
"resolution": "512,512",
"cache_latents": True,
"enable_bucket": False,
"min_bucket_reso": 256,
"max_bucket_reso": 1024,
"bucket_reso_steps": 64,
"bucket_no_upscale": False,
"reg_data_dir": None,
"in_json": "/tmp/rreid/rreid.json",
"dataset_repeats": 1,
"output_dir": "/mnt/d/Models/lora/",
"output_name": "lora-rreid-random-v1",
"save_precision": "fp16",
"save_every_n_epochs": 1,
"save_n_epoch_ratio": None,
"save_last_n_epochs": None,
"save_last_n_epochs_state": None,
"save_state": False,
"resume": None,
"train_batch_size": 1,
"max_token_length": None,
"use_8bit_adam": False,
"mem_eff_attn": False,
"xformers": False,
"vae": None,
"learning_rate": 1e-05,
"max_train_steps": 5000,
"max_train_epochs": None,
"max_data_loader_n_workers": 8,
"persistent_data_loader_workers": False,
"seed": 42,
"gradient_checkpointing": False,
"gradient_accumulation_steps": 1,
"mixed_precision": "fp16",
"full_fp16": False,
"clip_skip": None,
"logging_dir": None,
"log_prefix": None,
"lr_scheduler": "cosine",
"lr_warmup_steps": 0,
"prior_loss_weight": 1.0,
"no_metadata": False,
"save_model_as": "ckpt",
"unet_lr": 0.001,
"text_encoder_lr": 5e-05,
"lr_scheduler_num_cycles": 1,
"lr_scheduler_power": 1,
"network_weights": None,
"network_module": "networks.lora",
"network_dim": 16,
"network_alpha": 1.0,
"network_args": None,
"network_train_unet_only": False,
"network_train_text_encoder_only": False,
"training_comment": "mood-magic"
})
if __name__ == '__main__':
parser = argparse.ArgumentParser(description = 'train lora')
parser.add_argument('--model', type=str, default=None, required=True, help='original model to use a base for training')
parser.add_argument('--input', type=str, default=None, required=True, help='input folder with training images')
parser.add_argument('--dir', type=str, default=None, required=True, help='folder containing lora checkpoints')
parser.add_argument('--name', type=str, default=None, required=True, help='lora name')
parser.add_argument('--steps', type=int, default=5000, required=False, help='training steps')
parser.add_argument('--dim', type=int, default=16, required=False, help='network dimension')
parser.add_argument("--noprocess", default = False, action='store_true', help = "skip processing and use existing input data")
args = parser.parse_args()
if not os.path.exists(args.model) or not os.path.isfile(args.model):
log.error({ 'lora cannot find model': args.model })
exit(1)
options.pretrained_model_name_or_path = args.model
if not os.path.exists(args.input) or not os.path.isdir(args.input):
log.error({ 'lora cannot find training dir': args.input })
exit(1)
if not os.path.exists(args.dir) or not os.path.isdir(args.dir):
log.error({ 'lora cannot find training dir': args.dir })
exit(1)
options.output_dir = args.dir
options.output_name = args.name
options.max_train_steps = args.steps
options.network_dim = args.dim
log.info({ 'train lora args': vars(options) })
transformers.logging.set_verbosity_error()
if args.noprocess:
options.train_data_dir = args.input
else:
dir = os.path.join(tempfile.gettempdir(), args.name, '10_processed')
Path(dir).mkdir(parents=True, exist_ok=True)
files = []
json_data = {}
for root, _sub_dirs, folder in os.walk(args.input):
for f in folder:
files.append(os.path.join(root, f))
for f in files:
res = process_file(f = f, dst = dir, preview = False, offline = True)
log.info({ 'processed': res, 'inputs': len(files) })
options.train_data_dir = args.input
dir = os.path.join(tempfile.gettempdir(), args.name)
train(options)
"""
- cannot use `accelerate` with *dynamo* enabled
- cannot use `xformers` due to *faketensors* requirement
- cannot use `mem_eff_attn` due to *forwardfunc* mismatch
TODO
--gradient_checkpointing
--gradient_accumulation_steps=10
--caption_extension=txt
--in_json
WORKING
process.py --output "/tmp/rreid/img/10_processed" /home/vlado/generative/Input/ryanreid/random --offline
accelerate launch --no_python --quiet --num_cpu_threads_per_process=16 python /home/vlado/dev/automatic/modules/lora/train_network.py \
--pretrained_model_name_or_path="/mnt/d/Models/stable-diffusion/sd-v15-runwayml.ckpt" \
--train_data_dir="/tmp/rreid/img" \
--logging_dir="/tmp/rreid/logging" \
--output_dir="/mnt/d/Models/lora/" \
--output_name="lora-rreid-random-v1" \
--resolution=512,512 \
--learning_rate=1e-5 \
--unet_lr=1e-3 \
--text_encoder_lr=5e-5 \
--lr_scheduler_num_cycles=1 \
--lr_scheduler=cosine \
--max_train_steps=5000 \
--network_alpha=1 \
--network_dim=16 \
--network_module=networks.lora \
--save_every_n_epochs=1 \
--save_model_as=ckpt \
--save_precision=fp16 \
--mixed_precision=fp16 \
--seed=42 \
--train_batch_size=1 \
--cache_latents \
metadata { image_key: img_md: { caption: str, tags: [] } }
abs_path = glob_images(train_data_dir, image_key)
}}
./train-lora.py --model /mnt/d/Models/stable-diffusion/sd-v15-runwayml.ckpt --name rreid --dir /mnt/d/Models/lora --input ~/generative/Input/ryanreid/random/
"""
cli/modules/train-losschart.py
-2
View File
@@ -6,12 +6,10 @@ import sys
import json
import pathlib
import logging
import torch
import numpy as np
from PIL import Image, ImageFont, ImageDraw
from matplotlib import pyplot as plt
from util import log, Map
cli/modules/train-lossrate.py
-2
View File
@@ -5,11 +5,9 @@ auto-generate learn-rate
import io
import math
import logging
import numpy as np
from PIL import Image, ImageFont, ImageDraw
from matplotlib import pyplot as plt
from util import log, Map
cli/modules/video-extract.py
-1
View File
@@ -8,7 +8,6 @@ import subprocess
import pathlib
import argparse
import filetype
from util import log, Map