File size: 7,488 Bytes
8d82201 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 |
import os
import time
import copy
import logging
import math
import numpy as np
import torch
import random
import matplotlib.pyplot as plt
from detectron2.config import configurable
from detectron2.data import detection_utils as utils
from detectron2.data import transforms as T
from transformers import BertTokenizer
from pycocotools import mask as coco_mask
import albumentations as A
# from albumentations.pytorch import ToTensorV2
from PIL import Image, ImageDraw, ImageFilter
from detectron2.utils.visualizer import Visualizer
def convert_coco_poly_to_mask(segmentations, height, width):
masks = []
for polygons in segmentations:
rles = coco_mask.frPyObjects(polygons, height, width)
mask = coco_mask.decode(rles)
if len(mask.shape) < 3:
mask = mask[..., None]
mask = torch.as_tensor(mask, dtype=torch.uint8)
mask = mask.any(dim=2)
masks.append(mask)
if masks:
masks = torch.stack(masks, dim=0)
else:
masks = torch.zeros((0, height, width), dtype=torch.uint8)
return masks
def build_transform_train(cfg):
image_size = cfg.img_size
# min_scale = cfg.INPUT.MIN_SCALE
augmentation = []
augmentation.extend([
T.Resize((image_size, image_size))
])
return augmentation
def build_transform_test(cfg):
image_size = cfg.img_size
augmentation = []
augmentation.extend([
T.Resize((image_size, image_size))
])
return augmentation
def COCOVisualization(dataloader, dirname="coco-aug-data-vis"):
mean = (0.485, 0.456, 0.406)
std = (0.229, 0.224, 0.225)
denorm = A.Normalize(
mean=[-m / s for m, s in zip(mean, std)],
std=[1.0 / s for s in std],
max_pixel_value=1.0
)
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
sent_idx = 0
os.makedirs(dirname, exist_ok=True)
# dataloader = build_detection_train_loader(cfg, mapper=mapper)
it = iter(dataloader)
batch = next(it)
n_sample = random.randint(4, len(batch))
for i in range(n_sample):
batch = next(it)
img, gt_mask, lang_tokens, lang_mask = batch
img_np = np.transpose(img.cpu().numpy(), (1,2,0))
# img_denorm = denorm(image=img_np)['image']
# img_ndarray = (img_denorm*255).astype(np.uint8)
seg_target = gt_mask[:,:].cpu().numpy()
tokens = lang_tokens.reshape(-1).cpu().numpy()
sentences = tokenizer.decode(tokens, skip_special_tokens=True)
fpath = os.path.join(dirname, f'sample_{i+1}.jpg')
fig = plt.figure(figsize=(10,6))
ax1 = fig.add_subplot(1,2,1)
ax1.imshow(img_np.astype('uint8'))
ax1.set_xlabel("Mosaic Image")
ax2 = fig.add_subplot(1,2,2)
ax2.imshow(seg_target)
ax2.set_xlabel("Segmentation Map")
plt.suptitle(sentences)
plt.tight_layout()
plt.savefig(fpath)
# if 'gt_masks' in batch[0].keys():
# for i in range(n_sample):
# data = batch[i]
# img = data['image'].unsqueeze(0)
# img_np = np.transpose(img[0].cpu().numpy(), (1,2,0))
# img_denorm = denorm(image=img_np)['image']
# img_ndarray = (img_denorm*255).astype(np.uint8)
# seg_target = data['gt_masks'].squeeze(0)
# tensor_embedding = data['lang_tokens'][:,:]
# sentences = tokenizer.decode(tensor_embedding[0], skip_special_tokens=True)
# # tokens = [ds.tokenizer.decode([w], skip_special_tokens=False) for w in tensor_embedding[0]]
# # tokens = [x for x in tokens if x!='[PAD]']
# fpath = os.path.join(dirname, os.path.basename(data["file_name"]))
# fig = plt.figure(figsize=(10,6))
# ax1 = fig.add_subplot(1,2,1)
# ax1.imshow(img_ndarray)
# ax1.set_xlabel("Mosaic Image")
# ax2 = fig.add_subplot(1,2,2)
# ax2.imshow(seg_target)
# ax2.set_xlabel("Segmentation Map")
# plt.suptitle(sentences)
# plt.tight_layout()
# plt.savefig(fpath)
# else :
# for i in range(n_sample):
# d = batch[i]
# img = np.array(Image.open(d["file_name"]))
# visualizer = Visualizer(img, metadata={})
# vis = visualizer.draw_dataset_dict(d)
# fpath = os.path.join(dirname, os.path.basename(d["file_name"]))
# vis.save(fpath)
def MosaicVisualization(dataset, dirname="coco-aug-data-vis", n_sample=4):
mean = (0.485, 0.456, 0.406)
std = (0.229, 0.224, 0.225)
denorm = A.Normalize(
mean=[-m / s for m, s in zip(mean, std)],
std=[1.0 / s for s in std],
max_pixel_value=1.0
)
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
os.makedirs(dirname, exist_ok=True)
# dataset = build_detection_train_loader(cfg, mapper=mapper)
it = iter(dataset)
while(n_sample):
try :
data = next(it)
# n_sample = random.randint(1, len(batch))
# if 'seg_target' in batch[0].keys():
# for i in range(n_sample):
# data = batch[i]
img = data['image']
img_np = np.transpose(img.cpu().numpy(), (1,2,0))
img_denorm = denorm(image=img_np)['image']
img_ndarray = (img_denorm*255).astype(np.uint8)
seg_target = data['seg_target']
tensor_embedding = data['sentence'].reshape(-1).cpu().numpy()
sentences = tokenizer.decode(tensor_embedding, skip_special_tokens=True)
# tokens = [ds.tokenizer.decode([w], skip_special_tokens=False) for w in tensor_embedding[0]]
# tokens = [x for x in tokens if x!='[PAD]']
fpath = os.path.join(dirname, f'sample_{n_sample}.jpg')
fig = plt.figure(figsize=(10,6))
ax1 = fig.add_subplot(1,2,1)
ax1.imshow(img_ndarray)
ax1.set_xlabel("Mosaic Image")
ax2 = fig.add_subplot(1,2,2)
ax2.imshow(seg_target)
ax2.set_xlabel("Segmentation Map")
plt.suptitle(sentences)
plt.tight_layout()
plt.savefig(fpath)
n_sample -= 1
except :
break
# else :
# for i in range(n_sample):
# d = batch[i]
# img = np.array(Image.open(d["file_name"]))
# visualizer = Visualizer(img, metadata={})
# vis = visualizer.draw_dataset_dict(d)
# fpath = os.path.join(dirname, os.path.basename(d["file_name"]))
# vis.save(fpath)
def cosine_annealing(epoch, n_epochs, n_cycles, lrate_max=1):
"""
epoch : specific epoch you want to calcuate the probability
n_epochs : total number of epochs
n_cycles : number of cycle of cosine cf. 1 cycle = half of a period
lrate_max : maximum of probability
"""
epochs_per_cycle = math.floor(n_epochs/n_cycles)
cos_inner = (math.pi * (epoch % epochs_per_cycle)) / (epochs_per_cycle)
return lrate_max/2 * (math.cos(cos_inner) + 1)
def get_warmup_value(start_value, end_value, step, total_steps):
if step >= total_steps:
return end_value
mul = np.cos((1 - (step / total_steps)) * math.pi / 2) # Adjust the cosine function for warmup
warmup_range = end_value - start_value
return warmup_range * mul + start_value
|