Dev (CoinCheung#242)

* refactor dataset

* add customer dataset and associated config file

* add script of check dataset information

* modify script of check dataset

* format

* gitignore

.gitignore

@@ -104,8 +104,6 @@ venv.bak/
 
 
 ## Coin: 
-data
-data/
 play.py
 preprocess_data.py
 res/

README.md

@@ -109,7 +109,14 @@ frankfurt_000001_079206_leftImg8bit.png,frankfurt_000001_079206_gtFine_labelIds.
 ...
 ```
 Each line is a pair of training sample and ground truth image path, which are separated by a single comma `,`.   
-Then you need to change the field of `im_root` and `train/val_im_anns` in the configuration files. If you found what shows in `cityscapes_cv2.py` is not clear, you can also see `coco.py`.
+
+I recommand you to check the information of your dataset with the script:  
+```
+$ python tools/check_dataset_info.py --im_root /path/to/your/data_root --im_anns /path/to/your/anno_file
+```
+This will print some of the information of your dataset.  
+
+Then you need to change the field of `im_root` and `train/val_im_anns` in the config file. I prepared a demo config file for you named [`bisenet_customer.py`](./configs/bisenet_customer.py). You can start from this conig file.
 
 
 ## train

configs/bisenet_customer.py

@@ -0,0 +1,23 @@
+
+cfg = dict(
+    model_type='bisenetv1',
+    n_cats=20,
+    num_aux_heads=2,
+    lr_start=1e-2,
+    weight_decay=5e-4,
+    warmup_iters=1000,
+    max_iter=80000,
+    dataset='CustomerDataset',
+    im_root='./datasets/cityscapes',
+    train_im_anns='./datasets/cityscapes/train.txt',
+    val_im_anns='./datasets/cityscapes/val.txt',
+    scales=[0.75, 2.],
+    cropsize=[512, 512],
+    eval_crop=[512, 512],
+    eval_scales=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
+    ims_per_gpu=8,
+    eval_ims_per_gpu=2,
+    use_fp16=True,
+    use_sync_bn=False,
+    respth='./res',
+)

lib/data/__init__.py

@@ -0,0 +1,2 @@
+
+from .get_dataloader import get_data_loader

lib/base_dataset.py → lib/data/base_dataset.py

@@ -11,7 +11,6 @@
 import cv2
 import numpy as np
 
-from lib.sampler import RepeatedDistSampler
 
 
 

lib/cityscapes_cv2.py → lib/data/cityscapes_cv2.py

@@ -11,8 +11,8 @@
 import cv2
 import numpy as np
 
-import lib.transform_cv2 as T
-from lib.base_dataset import BaseDataset
+import lib.data.transform_cv2 as T
+from lib.data.base_dataset import BaseDataset
 
 
 labels_info = [

lib/coco.py → lib/data/coco.py

@@ -11,9 +11,8 @@
 import cv2
 import numpy as np
 
-import lib.transform_cv2 as T
-from lib.sampler import RepeatedDistSampler
-from lib.base_dataset import BaseDataset
+import lib.data.transform_cv2 as T
+from lib.data.base_dataset import BaseDataset
 
 '''
 91(thing) + 91(stuff) = 182 classes, label proportions are:
@@ -51,6 +50,7 @@ def __init__(self, dataroot, annpath, trans_func=None, mode='train'):
         super(CocoStuff, self).__init__(
                 dataroot, annpath, trans_func, mode)
         self.n_cats = 171 # 91 stuff, 91 thing, 11 of thing have no annos
+        self.lb_ignore = 255
 
         ## label mapping, remove non-existing labels
         missing = [11, 25, 28, 29, 44, 65, 67, 68, 70, 82, 90]

lib/data/customer_dataset.py

@@ -0,0 +1,21 @@
+#!/usr/bin/python
+# -*- encoding: utf-8 -*-
+
+
+import lib.data.transform_cv2 as T
+from lib.data.base_dataset import BaseDataset
+
+
+class CustomerDataset(BaseDataset):
+
+    def __init__(self, dataroot, annpath, trans_func=None, mode='train'):
+        super(CustomerDataset, self).__init__(
+                dataroot, annpath, trans_func, mode)
+        self.lb_ignore = 255
+
+        self.to_tensor = T.ToTensor(
+            mean=(0.4, 0.4, 0.4), # rgb
+            std=(0.2, 0.2, 0.2),
+        )
+
+

lib/get_dataloader.py → lib/data/get_dataloader.py

@@ -3,10 +3,12 @@
 from torch.utils.data import Dataset, DataLoader
 import torch.distributed as dist
 
-import lib.transform_cv2 as T
-from lib.sampler import RepeatedDistSampler
-from lib.cityscapes_cv2 import CityScapes
-from lib.coco import CocoStuff
+import lib.data.transform_cv2 as T
+from lib.data.sampler import RepeatedDistSampler
+
+from lib.data.cityscapes_cv2 import CityScapes
+from lib.data.coco import CocoStuff
+from lib.data.customer_dataset import CustomerDataset
 
 
 

tools/check_dataset_info.py

@@ -0,0 +1,121 @@
+
+import os
+import os.path as osp
+import argparse
+from tqdm import tqdm
+
+import cv2
+import numpy as np
+
+
+parse = argparse.ArgumentParser()
+parse.add_argument('--im_root', dest='im_root', type=str, default='./datasets/cityscapes',)
+parse.add_argument('--im_anns', dest='im_anns', type=str, default='./datasets/cityscapes/train.txt',)
+parse.add_argument('--lb_ignore', dest='lb_ignore', type=int, default=255)
+args = parse.parse_args()
+
+lb_ignore = args.lb_ignore
+
+
+with open(args.im_anns, 'r') as fr:
+    lines = fr.read().splitlines()
+
+n_pairs = len(lines)
+impaths, lbpaths = [], []
+for l in lines:
+    impth, lbpth = l.split(',')
+    impth = osp.join(args.im_root, impth)
+    lbpth = osp.join(args.im_root, lbpth)
+    impaths.append(impth)
+    lbpaths.append(lbpth)
+
+
+## shapes
+max_shape_area, min_shape_area = [0, 0], [100000, 100000]
+max_shape_height, min_shape_height = [0, 0], [100000, 100000]
+max_shape_width, min_shape_width = [0, 0], [100000, 100000]
+max_lb_val, min_lb_val = -1, 10000000
+for impth, lbpth in tqdm(zip(impaths, lbpaths), total=n_pairs):
+    im = cv2.imread(impth)[:, :, ::-1]
+    lb = cv2.imread(lbpth, 0)
+    assert im.shape[:2] == lb.shape
+
+    shape = lb.shape
+    area = shape[0] * shape[1]
+    if area > max_shape_area[0] * max_shape_area[1]:
+        max_shape_area = shape
+    if area < min_shape_area[0] * min_shape_area[1]:
+        min_shape_area = shape
+
+    if shape[0] > max_shape_height[0]:
+        max_shape_height = shape
+    if shape[0] < min_shape_height[0]:
+        min_shape_height = shape
+
+    if shape[1] > max_shape_width[1]:
+        max_shape_width = shape
+    if shape[1] < min_shape_width[1]:
+        min_shape_width = shape
+
+    lb = lb[lb != lb_ignore]
+    if lb.size > 0:
+        max_lb_val = max(max_lb_val, np.max(lb))
+        min_lb_val = min(min_lb_val, np.min(lb))
+
+min_lb_val = 0
+max_lb_val = 181
+lb_minlength = 182
+## label info
+lb_minlength = max_lb_val+1-min_lb_val
+lb_hist = np.zeros(lb_minlength)
+for lbpth in tqdm(lbpaths):
+    lb = cv2.imread(lbpth, 0)
+    lb = lb[lb != lb_ignore] + min_lb_val
+    lb_hist += np.bincount(lb, minlength=lb_minlength)
+
+lb_missing_vals = [ind + min_lb_val
+        for ind, el in enumerate(lb_hist.tolist()) if el == 0]
+lb_ratios = (lb_hist / lb_hist.sum()).tolist()
+
+
+## pixel mean/std
+rgb_mean = np.zeros(3).astype(np.float32)
+n_pixels = 0
+for impth in tqdm(impaths):
+    im = cv2.imread(impth)[:, :, ::-1].astype(np.float32)
+    im = im.reshape(-1, 3) / 255.
+    n_pixels += im.shape[0]
+    rgb_mean += im.sum(axis=0)
+rgb_mean = (rgb_mean / n_pixels)
+
+rgb_std = np.zeros(3).astype(np.float32)
+for impth in tqdm(impaths):
+    im = cv2.imread(impth)[:, :, ::-1].astype(np.float32)
+    im = im.reshape(-1, 3) / 255.
+
+    a = (im - rgb_mean.reshape(1, 3)) ** 2
+    rgb_std += a.sum(axis=0)
+rgb_std = (rgb_std / n_pixels) ** 0.5
+
+rgb_mean  = rgb_mean.tolist()
+rgb_std  = rgb_std.tolist()
+
+
+print('\n')
+print(f'there are {n_pairs} lines in {args.im_anns}, which means {n_pairs} image/label image pairs')
+print('\n')
+
+print(f'max and min image shapes by area are: {max_shape_area}, {min_shape_area}')
+print(f'max and min image shapes by height are: {max_shape_height}, {min_shape_height}')
+print(f'max and min image shapes by width are: {max_shape_width}, {min_shape_width}')
+print('\n')
+
+print(f'we ignore label value of {args.lb_ignore} in label images')
+print(f'label values are within range of [{min_lb_val}, {max_lb_val}]')
+print(f'label values that are missing: {lb_missing_vals}')
+print('ratios of each label value: ')
+print('\t', lb_ratios)
+print('\n')
+
+print('pixel mean rgb: ', rgb_mean)
+print('pixel std rgb: ', rgb_std)

tools/evaluate.py

@@ -22,7 +22,7 @@
 from lib.models import model_factory
 from configs import set_cfg_from_file
 from lib.logger import setup_logger
-from lib.get_dataloader import get_data_loader
+from lib.data import get_data_loader
 
 
 def get_round_size(size, divisor=32):

tools/train.py

@@ -19,7 +19,7 @@
 
 from lib.models import model_factory
 from configs import set_cfg_from_file
-from lib.get_dataloader import get_data_loader
+from lib.data import get_data_loader
 from tools.evaluate import eval_model
 from lib.ohem_ce_loss import OhemCELoss
 from lib.lr_scheduler import WarmupPolyLrScheduler

tools/train_amp.py

@@ -20,7 +20,7 @@
 
 from lib.models import model_factory
 from configs import set_cfg_from_file
-from lib.get_dataloader import get_data_loader
+from lib.data import get_data_loader
 from evaluate import eval_model
 from lib.ohem_ce_loss import OhemCELoss
 from lib.lr_scheduler import WarmupPolyLrScheduler