diff --git a/test.py b/test.py index b614eae6..a4bba9a3 100644 --- a/test.py +++ b/test.py @@ -3,6 +3,8 @@ import json import time from pathlib import Path +from torch.utils.data import DataLoader + from models import * from utils.datasets import * from utils.utils import * @@ -39,16 +41,21 @@ def test( model.to(device).eval() - # Get dataloader - # dataloader = torch.utils.data.DataLoader(LoadImagesAndLabels(test_path), batch_size=batch_size) - dataloader = LoadImagesAndLabels(test_path, batch_size=batch_size, img_size=img_size) + # Dataloader + dataset = LoadImagesAndLabels(test_path, img_size=img_size) + dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=0) mean_mAP, mean_R, mean_P, seen = 0.0, 0.0, 0.0, 0 print('%11s' * 5 % ('Image', 'Total', 'P', 'R', 'mAP')) mP, mR, mAPs, TP, jdict = [], [], [], [], [] AP_accum, AP_accum_count = np.zeros(nC), np.zeros(nC) coco91class = coco80_to_coco91_class() - for (imgs, targets, paths, shapes) in dataloader: + for imgs, targets, paths, shapes in dataloader: + # Unpad and collate targets + for j, t in enumerate(targets): + t[:, 0] = j + targets = torch.cat([t[t[:, 5].nonzero()] for t in targets], 0).squeeze(1) + targets = targets.to(device) t = time.time() output = model(imgs.to(device)) @@ -71,7 +78,7 @@ def test( if save_json: # [{"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}, ... box = detections[:, :4].clone() # xyxy - scale_coords(img_size, box, shapes[si]) # to original shape + scale_coords(img_size, box, (shapes[0][si], shapes[1][si])) # to original shape box = xyxy2xywh(box) # xywh box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner @@ -129,7 +136,7 @@ def test( # Print image mAP and running mean mAP print(('%11s%11s' + '%11.3g' * 4 + 's') % - (seen, dataloader.nF, mean_P, mean_R, mean_mAP, time.time() - t)) + (seen, len(dataset), mean_P, mean_R, mean_mAP, time.time() - t)) # Print mAP per class print('\nmAP Per Class:') @@ -139,7 +146,7 @@ def test( # Save JSON if save_json: - imgIds = [int(Path(x).stem.split('_')[-1]) for x in dataloader.img_files] + imgIds = [int(Path(x).stem.split('_')[-1]) for x in dataset.img_files] with open('results.json', 'w') as file: json.dump(jdict, file) diff --git a/train.py b/train.py index 4f621388..4406dfca 100644 --- a/train.py +++ b/train.py @@ -42,10 +42,8 @@ def train( optimizer = torch.optim.SGD(model.parameters(), lr=lr0, momentum=.9) # Dataloader - if num_workers > 0: - cv2.setNumThreads(0) # to prevent OpenCV from multithreading dataset = LoadImagesAndLabels(train_path, img_size=img_size, augment=True) - dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=num_workers) + dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=4) cutoff = -1 # backbone reaches to cutoff layer start_epoch = 0 @@ -103,17 +101,28 @@ def train( if int(name.split('.')[1]) < cutoff: # if layer < 75 p.requires_grad = False if (epoch == 0) else True - ui = -1 rloss = defaultdict(float) for i, (imgs, targets, _, _) in enumerate(dataloader): - if targets.shape[1] == 100: # multithreaded 100-size block - targets = targets.view((-1, 6)) - targets = targets[targets[:, 5].nonzero().squeeze()] + # Unpad and collate targets + for j, t in enumerate(targets): + t[:, 0] = j + targets = torch.cat([t[t[:, 5].nonzero()] for t in targets], 0).squeeze(1) - nT = targets.shape[0] + nT = len(targets) if nT == 0: # if no targets continue continue + # Plot images with bounding boxes + plot_images = False + if plot_images: + import matplotlib.pyplot as plt + plt.figure(figsize=(10, 10)) + for ip in range(batch_size): + labels = xywh2xyxy(targets[targets[:, 0] == ip, 2:6]).numpy() * img_size + plt.subplot(3, 3, ip + 1).imshow(imgs[ip].numpy().transpose(1, 2, 0)) + plt.plot(labels[:, [0, 2, 2, 0, 0]].T, labels[:, [1, 1, 3, 3, 1]].T, '.-') + plt.axis('off') + # SGD burn-in if (epoch == 0) and (i <= n_burnin): lr = lr0 * (i / n_burnin) ** 4 @@ -138,9 +147,8 @@ def train( optimizer.zero_grad() # Running epoch-means of tracked metrics - ui += 1 for key, val in loss_dict.items(): - rloss[key] = (rloss[key] * ui + val) / (ui + 1) + rloss[key] = (rloss[key] * i + val) / (i + 1) s = ('%8s%12s' + '%10.3g' * 7) % ( '%g/%g' % (epoch, epochs - 1), @@ -197,7 +205,7 @@ if __name__ == '__main__': parser.add_argument('--multi-scale', action='store_true', help='random image sizes per batch 320 - 608') parser.add_argument('--img-size', type=int, default=32 * 13, help='pixels') parser.add_argument('--resume', action='store_true', help='resume training flag') - parser.add_argument('--num_workers', type=int, default=4, help='number of Pytorch DataLoader workers') + parser.add_argument('--num_workers', type=int, default=0, help='number of Pytorch DataLoader workers') opt = parser.parse_args() print(opt, end='\n\n') diff --git a/utils/datasets.py b/utils/datasets.py index 90b67e54..6ccefcc3 100755 --- a/utils/datasets.py +++ b/utils/datasets.py @@ -6,6 +6,7 @@ import random import cv2 import numpy as np import torch +from torch.utils.data import Dataset from utils.utils import xyxy2xywh @@ -88,152 +89,102 @@ class LoadWebcam: # for inference return 0 -class LoadImagesAndLabels: # for training - def __init__(self, path, batch_size=1, img_size=608, augment=False): +class LoadImagesAndLabels(Dataset): # for training/testing + def __init__(self, path, img_size=416, augment=False): with open(path, 'r') as file: self.img_files = file.read().splitlines() self.img_files = list(filter(lambda x: len(x) > 0, self.img_files)) - - self.nF = len(self.img_files) # number of image files - self.nB = math.ceil(self.nF / batch_size) # number of batches - assert self.nF > 0, 'No images found in %s' % path - + assert len(self.img_files) > 0, 'No images found in %s' % path + self.img_size = img_size + self.augment = augment self.label_files = [x.replace('images', 'labels').replace('.png', '.txt').replace('.jpg', '.txt') for x in self.img_files] - self.batch_size = batch_size - self.img_size = img_size - self.augment = augment - - def __iter__(self): - self.count = -1 - #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF) - return self + def __len__(self): + return len(self.img_files) def __getitem__(self, index): - imgs, labels0, img_paths, img_shapes = self.load_images(index, index + 1) + img_path = self.img_files[index] + label_path = self.label_files[index] - labels0[:, 0] = index % self.batch_size - labels = torch.zeros(100, 6) - labels[:min(len(labels0), 100)] = labels0 # max 100 labels per image + img = cv2.imread(img_path) # BGR + assert img is not None, 'File Not Found ' + img_path - return imgs.squeeze(0), labels, img_paths, img_shapes + augment_hsv = True + if self.augment and augment_hsv: + # SV augmentation by 50% + fraction = 0.50 + img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) + S = img_hsv[:, :, 1].astype(np.float32) + V = img_hsv[:, :, 2].astype(np.float32) - def __next__(self): - self.count += 1 # batches - if self.count >= self.nB: - raise StopIteration + a = (random.random() * 2 - 1) * fraction + 1 + S *= a + if a > 1: + np.clip(S, a_min=0, a_max=255, out=S) - ia = self.count * self.batch_size # start index - ib = min(ia + self.batch_size, self.nF) # end index + a = (random.random() * 2 - 1) * fraction + 1 + V *= a + if a > 1: + np.clip(V, a_min=0, a_max=255, out=V) - return self.load_images(ia, ib) + img_hsv[:, :, 1] = S.astype(np.uint8) + img_hsv[:, :, 2] = V.astype(np.uint8) + cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img) - def load_images(self, ia, ib): - img_all, labels_all, img_paths, img_shapes = [], [], [], [] - for index, files_index in enumerate(range(ia, ib)): - img_path = self.img_files[files_index] - label_path = self.label_files[files_index] + h, w, _ = img.shape + img, ratio, padw, padh = letterbox(img, height=self.img_size) - img = cv2.imread(img_path) # BGR - assert img is not None, 'File Not Found ' + img_path + # Load labels + if os.path.isfile(label_path): + # labels0 = np.loadtxt(label_path, dtype=np.float32).reshape(-1, 5) # SLOWER + with open(label_path, 'r') as file: + lines = file.read().splitlines() + labels0 = np.array([x.split() for x in lines], dtype=np.float32) - augment_hsv = True - if self.augment and augment_hsv: - # SV augmentation by 50% - fraction = 0.50 - img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) - S = img_hsv[:, :, 1].astype(np.float32) - V = img_hsv[:, :, 2].astype(np.float32) + # Normalized xywh to pixel xyxy format + labels = labels0.copy() + labels[:, 1] = ratio * w * (labels0[:, 1] - labels0[:, 3] / 2) + padw + labels[:, 2] = ratio * h * (labels0[:, 2] - labels0[:, 4] / 2) + padh + labels[:, 3] = ratio * w * (labels0[:, 1] + labels0[:, 3] / 2) + padw + labels[:, 4] = ratio * h * (labels0[:, 2] + labels0[:, 4] / 2) + padh + else: + labels = np.array([]) - a = (random.random() * 2 - 1) * fraction + 1 - S *= a - if a > 1: - np.clip(S, a_min=0, a_max=255, out=S) + # Augment image and labels + if self.augment: + img, labels, M = random_affine(img, labels, degrees=(-5, 5), translate=(0.10, 0.10), scale=(0.90, 1.10)) - a = (random.random() * 2 - 1) * fraction + 1 - V *= a - if a > 1: - np.clip(V, a_min=0, a_max=255, out=V) + nL = len(labels) + if nL > 0: + # convert xyxy to xywh + labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / self.img_size - img_hsv[:, :, 1] = S.astype(np.uint8) - img_hsv[:, :, 2] = V.astype(np.uint8) - cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img) + if self.augment: + # random left-right flip + lr_flip = True + if lr_flip & (random.random() > 0.5): + img = np.fliplr(img) + if nL > 0: + labels[:, 1] = 1 - labels[:, 1] - h, w, _ = img.shape - img, ratio, padw, padh = letterbox(img, height=self.img_size) + # random up-down flip + ud_flip = False + if ud_flip & (random.random() > 0.5): + img = np.flipud(img) + if nL > 0: + labels[:, 2] = 1 - labels[:, 2] - # Load labels - if os.path.isfile(label_path): - # labels0 = np.loadtxt(label_path, dtype=np.float32).reshape(-1, 5) # SLOWER - with open(label_path, 'r') as file: - lines = file.read().splitlines() - labels0 = np.array([x.split() for x in lines], dtype=np.float32) - - # Normalized xywh to pixel xyxy format - labels = labels0.copy() - labels[:, 1] = ratio * w * (labels0[:, 1] - labels0[:, 3] / 2) + padw - labels[:, 2] = ratio * h * (labels0[:, 2] - labels0[:, 4] / 2) + padh - labels[:, 3] = ratio * w * (labels0[:, 1] + labels0[:, 3] / 2) + padw - labels[:, 4] = ratio * h * (labels0[:, 2] + labels0[:, 4] / 2) + padh - else: - labels = np.array([]) - - # Augment image and labels - if self.augment: - img, labels, M = random_affine(img, labels, degrees=(-5, 5), translate=(0.10, 0.10), scale=(0.90, 1.10)) - - plotFlag = False - if plotFlag: - import matplotlib.pyplot as plt - plt.figure(figsize=(10, 10)) if index == 0 else None - plt.subplot(4, 4, index + 1).imshow(img[:, :, ::-1]) - plt.plot(labels[:, [1, 3, 3, 1, 1]].T, labels[:, [2, 2, 4, 4, 2]].T, '.-') - plt.axis('off') - - nL = len(labels) - if nL > 0: - # convert xyxy to xywh - labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / self.img_size - - if self.augment: - # random left-right flip - lr_flip = True - if lr_flip & (random.random() > 0.5): - img = np.fliplr(img) - if nL > 0: - labels[:, 1] = 1 - labels[:, 1] - - # random up-down flip - ud_flip = False - if ud_flip & (random.random() > 0.5): - img = np.flipud(img) - if nL > 0: - labels[:, 2] = 1 - labels[:, 2] - - if nL > 0: - labels = np.concatenate((np.zeros((nL, 1), dtype='float32') + index, labels), 1) - labels_all.append(labels) - - img_all.append(img) - img_paths.append(img_path) - img_shapes.append((h, w)) + labels_out = np.zeros((100, 6), dtype=np.float32) + if nL > 0: + labels_out[:nL, 1:] = labels # max 100 labels per image # Normalize - img_all = np.stack(img_all)[:, :, :, ::-1].transpose(0, 3, 1, 2) # list to np.array and BGR to RGB - img_all = np.ascontiguousarray(img_all, dtype=np.float32) # uint8 to float32 - img_all /= 255.0 # 0 - 255 to 0.0 - 1.0 + img = img[:, :, ::-1].transpose(2, 0, 1) # list to np.array and BGR to RGB + img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32 + img /= 255.0 # 0 - 255 to 0.0 - 1.0 - if len(labels_all) > 0: - labels_all = np.concatenate(labels_all, 0) - else: - labels_all = np.zeros((1, 6), dtype='float32') - - labels_all = torch.from_numpy(labels_all) - return torch.from_numpy(img_all), labels_all, img_paths, img_shapes - - def __len__(self): - return self.nB # number of batches + return torch.from_numpy(img), torch.from_numpy(labels_out), img_path, (h, w) def letterbox(img, height=416, color=(127.5, 127.5, 127.5)): # resize a rectangular image to a padded square diff --git a/utils/utils.py b/utils/utils.py index de79eac6..1f4e03b0 100755 --- a/utils/utils.py +++ b/utils/utils.py @@ -15,6 +15,9 @@ from utils import torch_utils torch.set_printoptions(linewidth=1320, precision=5, profile='long') np.set_printoptions(linewidth=320, formatter={'float_kind': '{:11.5g}'.format}) # format short g, %precision=5 +# Prevent OpenCV from multithreading (to use PyTorch DataLoader) +cv2.setNumThreads(0) + def float3(x): # format floats to 3 decimals return float(format(x, '.3f'))