stable-diffusion-webui-dump.../scripts/dumpunet/features/process.py

191 lines
5.3 KiB
Python

import os
import math
from typing import Generator
from torch import Tensor
import numpy as np
from PIL import Image
from modules import shared
from scripts.dumpunet import layerinfo
from scripts.dumpunet.features.featureinfo import FeatureInfo, Features, MultiImageFeatures
from scripts.dumpunet.report import message as E
def feature_diff(
features1: MultiImageFeatures,
features2: MultiImageFeatures
) -> Generator[tuple[int,int,str,Tensor],None,None]:
# features1 and features2 must be have same keys...
for img_idx in sorted(features1.keys()):
assert img_idx in features1
assert img_idx in features2
fs1 = features1[img_idx]
fs2 = features2[img_idx]
for step in sorted(fs1.keys()):
assert step in fs1
assert step in fs2
f1 = fs1[step]
f2 = fs2[step]
for layer in f1.layers():
assert layer in f1
assert layer in f2
l1, l2 = f1[layer], f2[layer]
a, b = l1.output, l2.output
assert a.size() == b.size()
assert len(a.size()) == 3
c = (b - a).abs()
yield img_idx, step, layer, c
def feature_to_grid_images(
feature: FeatureInfo,
layer: str,
width: int,
height: int,
color: bool
):
return tensor_to_grid_images(feature.output, layer, width, height, color)
def tensor_to_grid_images(
tensor: Tensor,
layer: str,
width: int,
height: int,
color: bool
):
assert len(tensor.size()) == 3
grid_x, grid_y = _get_grid_num(layer, width, height)
canvases = tensor_to_image(tensor, grid_x, grid_y, color)
return canvases
def save_features(
feature: FeatureInfo,
save_dir: str,
basename: str
):
save_tensor(feature.output, save_dir, basename)
def tensor_to_image(
tensor: Tensor,
grid_x: int,
grid_y: int,
color: bool,
):
# Regardless of wheather --opt-channelslast is enabled or not,
# feature.size() seems to return (batch, ch, h, w).
# Is this intended behaviour???
assert len(tensor.size()) == 3
max_ch, ih, iw = tensor.size()
width = (grid_x * (iw + 1) - 1)
height = (grid_y * (ih + 1) - 1)
def each_slice(it: range, n: int):
cur = []
for x in it:
cur.append(x)
if n == len(cur):
yield cur
cur = []
if 0 < len(cur):
yield cur
canvases: list[Image.Image] = []
color_format = "RGB" if color else "L"
for chs in each_slice(range(max_ch), grid_x * grid_y):
chs = list(chs)
canvas = Image.new(color_format, (width, height), color=0)
for iy in range(grid_y):
if len(chs) == 0:
break
for ix in range(grid_x):
if shared.state.interrupted:
break
if len(chs) == 0:
break
ch = chs.pop(0)
array = tensor[ch].cpu().numpy().astype(np.float32)
# create image
x = (iw+1) * ix
y = (ih+1) * iy
image = _tensor_to_image(array, color)
canvas.paste(Image.fromarray(image, color_format), (x, y))
canvases.append(canvas)
return canvases
def save_tensor(
tensor: Tensor,
save_dir: str,
basename: str
):
assert len(tensor.size()) == 3
for ch, t in enumerate(tensor):
filename = basename.format(ch=ch)
binpath = os.path.join(save_dir, filename + ".bin")
with open(binpath, "wb") as io:
array = t.cpu().numpy().astype(np.float32)
io.write(bytearray(array))
def _tensor_to_image(array: np.ndarray, color: bool):
# array := (-∞, ∞)
if color:
def colorize(v: float):
# v = -1 .. 1 を
# v < 0 のとき青 (0, 0, 1)
# v > 0 のとき赤 (1 ,0, 0)
# にする
rgb = (v if v > 0.0 else 0.0, 0.0, -v if v < 0.0 else 0.0)
return rgb
colorize2 = np.vectorize(colorize, otypes=[np.float32, np.float32, np.float32])
rgb = colorize2(np.clip(array, -1.0, 1.0))
return np.clip((np.dstack(rgb) * 256), 0, 255).astype(np.uint8)
else:
return np.clip(np.abs(array) * 256, 0, 255).astype(np.uint8)
def _get_grid_num(layer: str, width: int, height: int):
assert layer is not None and layer != "", E("<Layers> must not be empty.")
assert layer in layerinfo.Settings, E(f"Invalid <Layers> value: {layer}.")
_, (ch, mh, mw) = layerinfo.Settings[layer]
iw = math.ceil(width / 64)
ih = math.ceil(height / 64)
w = mw * iw
h = mh * ih
# w : width of a feature map
# h : height of a feature map
# ch: a number of a feature map
n = [w, h]
while ch % 2 == 0:
n[n[0]>n[1]] *= 2
ch //= 2
n[n[0]>n[1]] *= ch
if n[0] > n[1]:
while n[0] > n[1] * 2 and (n[0] // w) % 2 == 0:
n[0] //= 2
n[1] *= 2
else:
while n[0] * 2 < n[1] and (n[1] // h) % 2 == 0:
n[0] *= 2
n[1] //= 2
return [n[0] // w, n[1] // h]