p83571920
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

remove dependency of mmcv

This commit is contained in:
zhuyuxiao 2021-09-10 15:45:25 +08:00
parent f3769b1b58
commit 35a7af4a4d
7 changed files with 154 additions and 65 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
model_zoo/official/cv/ctpn/src/dataset.py
View File

@ -17,7 +17,7 @@
from __future__ import division
import numpy as np
from numpy import random
import mmcv
import cv2
import mindspore.dataset as de
import mindspore.dataset.vision.c_transforms as C
import mindspore.dataset.transforms.c_transforms as CC
@ -50,7 +50,7 @@ class PhotoMetricDistortion:
self.contrast_upper)
img *= alpha
# convert color from BGR to HSV
img = mmcv.bgr2hsv(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_BGR2HSV')))
# random saturation
if random.randint(2):
img[..., 1] *= random.uniform(self.saturation_lower,
@ -61,7 +61,7 @@ class PhotoMetricDistortion:
img[..., 0][img[..., 0] > 360] -= 360
img[..., 0][img[..., 0] < 0] += 360
# convert color from HSV to BGR
img = mmcv.hsv2bgr(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_HSV2BGR')))
# random contrast
if mode == 0:
if random.randint(2):
@ -97,11 +97,19 @@ class Expand:
boxes += np.tile((left, top), 2)
return img, boxes, labels
def rescale_with_tuple(img, scale):
h, w = img.shape[:2]
scale_factor = min(max(scale) / max(h, w), min(scale) / min(h, w))
new_size = int(w * float(scale_factor) + 0.5), int(h * float(scale_factor) + 0.5)
rescaled_img = cv2.resize(img, new_size, interpolation=cv2.INTER_LINEAR)
return rescaled_img, scale_factor
def rescale_column(img, gt_bboxes, gt_label, gt_num, img_shape):
"""rescale operation for image"""
img_data, scale_factor = mmcv.imrescale(img, (config.img_width, config.img_height), return_scale=True)
img_data, scale_factor = rescale_with_tuple(img, (config.img_width, config.img_height))
if img_data.shape[0] > config.img_height:
img_data, scale_factor2 = mmcv.imrescale(img_data, (config.img_height, config.img_width), return_scale=True)
img_data, scale_factor2 = rescale_with_tuple(img_data, (config.img_height, config.img_width))
scale_factor = scale_factor * scale_factor2
img_shape = np.append(img_shape, scale_factor)
img_shape = np.asarray(img_shape, dtype=np.float32)
@ -117,8 +125,12 @@ def rescale_column(img, gt_bboxes, gt_label, gt_num, img_shape):
def resize_column(img, gt_bboxes, gt_label, gt_num, img_shape):
"""resize operation for image"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width, 1.0)
@ -134,8 +146,12 @@ def resize_column(img, gt_bboxes, gt_label, gt_num, img_shape):
def resize_column_test(img, gt_bboxes, gt_label, gt_num, img_shape):
"""resize operation for image of eval"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width)

6
model_zoo/official/cv/deeptext/postprocess.py
View File

@ -17,7 +17,7 @@
import os
from PIL import Image
import numpy as np
import mmcv
import cv2
from src.utils import metrics
from model_utils.config import config
@ -37,8 +37,8 @@ def get_pred(file, result_path):
def get_gt_bboxes_labels(label_file, img_file):
img_data = np.array(Image.open(img_file))
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data, w_scale, h_scale = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width, 1.0)

29
model_zoo/official/cv/deeptext/src/dataset.py
View File

@ -20,7 +20,7 @@ import os
import numpy as np
from numpy import random
import mmcv
import cv2
import mindspore.dataset as de
import mindspore.dataset.vision.c_transforms as C
import mindspore.dataset.transforms.c_transforms as CC
@ -107,7 +107,7 @@ class PhotoMetricDistortion:
img *= alpha
# convert color from BGR to HSV
img = mmcv.bgr2hsv(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_BGR2HSV')))
# random saturation
if random.randint(2):
@ -121,7 +121,7 @@ class PhotoMetricDistortion:
img[..., 0][img[..., 0] < 0] += 360
# convert color from HSV to BGR
img = mmcv.hsv2bgr(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_HSV2BGR')))
# random contrast
if mode == 0:
@ -166,8 +166,8 @@ class Expand:
def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""resize operation for image"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data, w_scale, h_scale = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width, 1.0)
@ -184,8 +184,8 @@ def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num):
"""resize operation for image of eval"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data, w_scale, h_scale = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width)
@ -202,14 +202,23 @@ def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num):
def impad_to_multiple_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""impad operation for image"""
img_data = mmcv.impad(img, (config.img_height, config.img_width))
img_data = cv2.copyMakeBorder(img,
0, config.img_height - img.shape[0], 0, config.img_width - img.shape[1],
cv2.BORDER_CONSTANT,
value=0)
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num)
def imnormalize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""imnormalize operation for image"""
img_data = mmcv.imnormalize(img, [123.675, 116.28, 103.53], [58.395, 57.12, 57.375], True)
mean = np.asarray([123.675, 116.28, 103.53])
std = np.asarray([58.395, 57.12, 57.375])
img_data = img.copy().astype(np.float32)
cv2.cvtColor(img_data, cv2.COLOR_BGR2RGB, img_data) # inplace
cv2.subtract(img_data, np.float64(mean.reshape(1, -1)), img_data) # inplace
cv2.multiply(img_data, 1 / np.float64(std.reshape(1, -1)), img_data) # inplace
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num)
@ -217,7 +226,7 @@ def imnormalize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
def flip_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""flip operation for image"""
img_data = img
img_data = mmcv.imflip(img_data)
img_data = np.flip(img_data, axis=1)
flipped = gt_bboxes.copy()
_, w, _ = img_data.shape

56
model_zoo/official/cv/faster_rcnn/src/dataset.py
View File

@ -21,7 +21,6 @@ import numpy as np
from numpy import random
import cv2
import mmcv
import mindspore.dataset as de
import mindspore.dataset.vision.c_transforms as C
from mindspore.mindrecord import FileWriter
@ -103,7 +102,7 @@ class PhotoMetricDistortion:
img *= alpha
# convert color from BGR to HSV
img = mmcv.bgr2hsv(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_BGR2HSV')))
# random saturation
if random.randint(2):
@ -117,7 +116,7 @@ class PhotoMetricDistortion:
img[..., 0][img[..., 0] < 0] += 360
# convert color from HSV to BGR
img = mmcv.hsv2bgr(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_HSV2BGR')))
# random contrast
if mode == 0:
@ -157,12 +156,24 @@ class Expand:
boxes += np.tile((left, top), 2)
return img, boxes, labels
def rescale_with_tuple(img, scale):
h, w = img.shape[:2]
scale_factor = min(max(scale) / max(h, w), min(scale) / min(h, w))
new_size = int(w * float(scale_factor) + 0.5), int(h * float(scale_factor) + 0.5)
rescaled_img = cv2.resize(img, new_size, interpolation=cv2.INTER_LINEAR)
return rescaled_img, scale_factor
def rescale_with_factor(img, scale_factor):
h, w = img.shape[:2]
new_size = int(w * float(scale_factor) + 0.5), int(h * float(scale_factor) + 0.5)
return cv2.resize(img, new_size, interpolation=cv2.INTER_NEAREST)
def rescale_column(img, img_shape, gt_bboxes, gt_label, gt_num, config):
"""rescale operation for image"""
img_data, scale_factor = mmcv.imrescale(img, (config.img_width, config.img_height), return_scale=True)
img_data, scale_factor = rescale_with_tuple(img, (config.img_width, config.img_height))
if img_data.shape[0] > config.img_height:
img_data, scale_factor2 = mmcv.imrescale(img_data, (config.img_height, config.img_height), return_scale=True)
img_data, scale_factor2 = rescale_with_tuple(img_data, (config.img_height, config.img_height))
scale_factor = scale_factor*scale_factor2
gt_bboxes = gt_bboxes * scale_factor
@ -183,9 +194,9 @@ def rescale_column(img, img_shape, gt_bboxes, gt_label, gt_num, config):
def rescale_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, config):
"""rescale operation for image of eval"""
img_data, scale_factor = mmcv.imrescale(img, (config.img_width, config.img_height), return_scale=True)
img_data, scale_factor = rescale_with_tuple(img, (config.img_width, config.img_height))
if img_data.shape[0] > config.img_height:
img_data, scale_factor2 = mmcv.imrescale(img_data, (config.img_height, config.img_height), return_scale=True)
img_data, scale_factor2 = rescale_with_tuple(img_data, (config.img_height, config.img_height))
scale_factor = scale_factor*scale_factor2
pad_h = config.img_height - img_data.shape[0]
@ -204,8 +215,12 @@ def rescale_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, config):
def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num, config):
"""resize operation for image"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width, 1.0)
@ -222,8 +237,12 @@ def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num, config):
def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, config):
"""resize operation for image of eval"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = np.append(img_shape, (h_scale, w_scale))
@ -239,14 +258,23 @@ def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, config):
def impad_to_multiple_column(img, img_shape, gt_bboxes, gt_label, gt_num, config):
"""impad operation for image"""
img_data = mmcv.impad(img, (config.img_height, config.img_width))
img_data = cv2.copyMakeBorder(img,
0, config.img_height - img.shape[0], 0, config.img_width - img.shape[1],
cv2.BORDER_CONSTANT,
value=0)
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num)
def imnormalize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""imnormalize operation for image"""
img_data = mmcv.imnormalize(img, np.array([123.675, 116.28, 103.53]), np.array([58.395, 57.12, 57.375]), True)
mean = np.asarray([123.675, 116.28, 103.53])
std = np.asarray([58.395, 57.12, 57.375])
img_data = img.copy().astype(np.float32)
cv2.cvtColor(img_data, cv2.COLOR_BGR2RGB, img_data) # inplace
cv2.subtract(img_data, np.float64(mean.reshape(1, -1)), img_data) # inplace
cv2.multiply(img_data, 1 / np.float64(std.reshape(1, -1)), img_data) # inplace
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num)
@ -254,7 +282,7 @@ def imnormalize_column(img, img_shape, gt_bboxes, gt_label, gt_num):
def flip_column(img, img_shape, gt_bboxes, gt_label, gt_num):
"""flip operation for image"""
img_data = img
img_data = mmcv.imflip(img_data)
img_data = np.flip(img_data, axis=1)
flipped = gt_bboxes.copy()
_, w, _ = img_data.shape

15
model_zoo/official/cv/faster_rcnn/src/util.py
View File

@ -15,7 +15,6 @@
"""coco eval for fasterrcnn"""
import json
import numpy as np
import mmcv
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
@ -42,7 +41,7 @@ def coco_eval(result_files, result_types, coco, max_dets=(100, 300, 1000), singl
if not anns:
return summary_init
if mmcv.is_str(coco):
if isinstance(coco, str):
coco = COCO(coco)
assert isinstance(coco, COCO)
@ -208,18 +207,22 @@ def results2json(dataset, results, out_file):
json_results = det2json(dataset, results)
result_files['bbox'] = '{}.{}.json'.format(out_file, 'bbox')
result_files['proposal'] = '{}.{}.json'.format(out_file, 'bbox')
mmcv.dump(json_results, result_files['bbox'])
with open(result_files['bbox'], 'w') as fp:
json.dump(json_results, fp)
elif isinstance(results[0], tuple):
json_results = segm2json(dataset, results)
result_files['bbox'] = '{}.{}.json'.format(out_file, 'bbox')
result_files['proposal'] = '{}.{}.json'.format(out_file, 'bbox')
result_files['segm'] = '{}.{}.json'.format(out_file, 'segm')
mmcv.dump(json_results[0], result_files['bbox'])
mmcv.dump(json_results[1], result_files['segm'])
with open(result_files['bbox'], 'w') as fp:
json.dump(json_results[0], fp)
with open(result_files['segm'], 'w') as fp:
json.dump(json_results[1], fp)
elif isinstance(results[0], np.ndarray):
json_results = proposal2json(dataset, results)
result_files['proposal'] = '{}.{}.json'.format(out_file, 'proposal')
mmcv.dump(json_results, result_files['proposal'])
with open(result_files['proposal'], 'w') as fp:
json.dump(json_results, fp)
else:
raise TypeError('invalid type of results')
return result_files

61
model_zoo/official/cv/maskrcnn_mobilenetv1/src/dataset.py
View File

@ -18,7 +18,6 @@ from __future__ import division
import os
import re
import mmcv
import cv2
import numpy as np
from numpy import random
@ -117,7 +116,7 @@ class PhotoMetricDistortion:
img *= alpha
# convert color from BGR to HSV
img = mmcv.bgr2hsv(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_BGR2HSV')))
# random saturation
if random.randint(2):
@ -131,7 +130,7 @@ class PhotoMetricDistortion:
img[..., 0][img[..., 0] < 0] += 360
# convert color from HSV to BGR
img = mmcv.hsv2bgr(img)
img = cv2.cvtColor((img, getattr(cv2, f'COLOR_HSV2BGR')))
# random contrast
if mode == 0:
@ -176,11 +175,24 @@ class Expand:
return img, boxes, labels, mask
def rescale_with_tuple(img, scale):
h, w = img.shape[:2]
scale_factor = min(max(scale) / max(h, w), min(scale) / min(h, w))
new_size = int(w * float(scale_factor) + 0.5), int(h * float(scale_factor) + 0.5)
rescaled_img = cv2.resize(img, new_size, interpolation=cv2.INTER_LINEAR)
return rescaled_img, scale_factor
def rescale_with_factor(img, scale_factor):
h, w = img.shape[:2]
new_size = int(w * float(scale_factor) + 0.5), int(h * float(scale_factor) + 0.5)
return cv2.resize(img, new_size, interpolation=cv2.INTER_NEAREST)
def rescale_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""rescale operation for image"""
img_data, scale_factor = mmcv.imrescale(img, (config.img_width, config.img_height), return_scale=True)
img_data, scale_factor = rescale_with_tuple(img, (config.img_width, config.img_height))
if img_data.shape[0] > config.img_height:
img_data, scale_factor2 = mmcv.imrescale(img_data, (config.img_height, config.img_height), return_scale=True)
img_data, scale_factor2 = rescale_with_tuple(img_data, (config.img_height, config.img_height))
scale_factor = scale_factor*scale_factor2
gt_bboxes = gt_bboxes * scale_factor
@ -188,7 +200,7 @@ def rescale_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
gt_bboxes[:, 1::2] = np.clip(gt_bboxes[:, 1::2], 0, img_data.shape[0] - 1)
gt_mask_data = np.array([
mmcv.imrescale(mask, scale_factor, interpolation='nearest')
rescale_with_factor(mask, scale_factor)
for mask in gt_mask
])
@ -210,9 +222,9 @@ def rescale_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
def rescale_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""rescale operation for image of eval"""
img_data, scale_factor = mmcv.imrescale(img, (config.img_width, config.img_height), return_scale=True)
img_data, scale_factor = rescale_with_tuple(img, (config.img_width, config.img_height))
if img_data.shape[0] > config.img_height:
img_data, scale_factor2 = mmcv.imrescale(img_data, (config.img_height, config.img_height), return_scale=True)
img_data, scale_factor2 = rescale_with_tuple(img_data, (config.img_height, config.img_height))
scale_factor = scale_factor*scale_factor2
pad_h = config.img_height - img_data.shape[0]
@ -230,8 +242,12 @@ def rescale_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""resize operation for image"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
scale_factor = np.array(
[w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
img_shape = (config.img_height, config.img_width, 1.0)
@ -242,7 +258,7 @@ def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
gt_bboxes[:, 1::2] = np.clip(gt_bboxes[:, 1::2], 0, img_shape[0] - 1) # y1, y2 [0, H-1]
gt_mask_data = np.array([
mmcv.imresize(mask, (config.img_width, config.img_height), interpolation='nearest')
cv2.resize(mask, (config.img_width, config.img_height), interpolation=cv2.INTER_NEAREST)
for mask in gt_mask
])
return (img_data, img_shape, gt_bboxes, gt_label, gt_num, gt_mask_data)
@ -250,8 +266,12 @@ def resize_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""resize operation for image of eval"""
img_data = img
img_data, w_scale, h_scale = mmcv.imresize(
img_data, (config.img_width, config.img_height), return_scale=True)
img_data, w_scale, h_scale = cv2.resize(
img_data, (config.img_width, config.img_height), interpolation=cv2.INTER_LINEAR)
h, w = img_data.shape[:2]
h_scale = config.img_height / h
w_scale = config.img_width / w
img_shape = np.append(img_shape, (h_scale, w_scale))
img_shape = np.asarray(img_shape, dtype=np.float32)
@ -259,20 +279,29 @@ def resize_column_test(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
def impad_to_multiple_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""impad operation for image"""
img_data = mmcv.impad(img, (config.img_height, config.img_width))
img_data = cv2.copyMakeBorder(img,
0, config.img_height - img.shape[0], 0, config.img_width - img.shape[1],
cv2.BORDER_CONSTANT,
value=0)
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num, gt_mask)
def imnormalize_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""imnormalize operation for image"""
img_data = mmcv.imnormalize(img, np.array([123.675, 116.28, 103.53]), np.array([58.395, 57.12, 57.375]), True)
mean = np.asarray([123.675, 116.28, 103.53])
std = np.asarray([58.395, 57.12, 57.375])
img_data = img.copy().astype(np.float32)
cv2.cvtColor(img_data, cv2.COLOR_BGR2RGB, img_data) # inplace
cv2.subtract(img_data, np.float64(mean.reshape(1, -1)), img_data) # inplace
cv2.multiply(img_data, 1 / np.float64(std.reshape(1, -1)), img_data) # inplace
img_data = img_data.astype(np.float32)
return (img_data, img_shape, gt_bboxes, gt_label, gt_num, gt_mask)
def flip_column(img, img_shape, gt_bboxes, gt_label, gt_num, gt_mask):
"""flip operation for image"""
img_data = img
img_data = mmcv.imflip(img_data)
img_data = np.flip(img_data, axis=1)
flipped = gt_bboxes.copy()
_, w, _ = img_data.shape

18
model_zoo/official/cv/maskrcnn_mobilenetv1/src/util.py
View File

@ -14,7 +14,7 @@
# ============================================================================
"""coco eval for maskrcnn_mobilenetv1"""
import json
import mmcv
import cv2
import numpy as np
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
@ -44,7 +44,7 @@ def coco_eval(result_files, result_types, coco, max_dets=(100, 300, 1000), singl
anns = json.load(open(result_files['bbox']))
if not anns:
return summary_init
if mmcv.is_str(coco):
if isinstance(coco, str):
coco = COCO(coco)
assert isinstance(coco, COCO)
@ -216,17 +216,21 @@ def results2json(dataset, results, out_file):
json_results = det2json(dataset, results)
result_files['bbox'] = '{}.{}.json'.format(out_file, 'bbox')
result_files['proposal'] = '{}.{}.json'.format(out_file, 'bbox')
mmcv.dump(json_results, result_files['bbox'])
with open(result_files['bbox'], 'w') as fp:
json.dump(json_results, fp)
elif isinstance(results[0], tuple):
json_results = segm2json(dataset, results)
result_files['bbox'] = '{}.{}.json'.format(out_file, 'bbox')
result_files['segm'] = '{}.{}.json'.format(out_file, 'segm')
mmcv.dump(json_results[0], result_files['bbox'])
mmcv.dump(json_results[1], result_files['segm'])
with open(result_files['bbox'], 'w') as fp:
json.dump(json_results[0], fp)
with open(result_files['segm'], 'w') as fp:
json.dump(json_results[1], fp)
elif isinstance(results[0], np.ndarray):
json_results = proposal2json(dataset, results)
result_files['proposal'] = '{}.{}.json'.format(out_file, 'proposal')
mmcv.dump(json_results, result_files['proposal'])
with open(result_files['proposal'], 'w') as fp:
json.dump(json_results, fp)
else:
raise TypeError('invalid type of results')
return result_files
@ -261,7 +265,7 @@ def get_seg_masks(mask_pred, det_bboxes, det_labels, img_meta, rescale, num_clas
h = max(h, 1)
mask_pred_ = mask_pred[i, :, :]
im_mask = np.zeros((img_h, img_w), dtype=np.uint8)
bbox_mask = mmcv.imresize(mask_pred_, (w, h))
bbox_mask = cv2.resize(mask_pred_, (w, h), interpolation=cv2.INTER_LINEAR)
bbox_mask = (bbox_mask > config.mask_thr_binary).astype(np.uint8)
im_mask[bbox[1]:bbox[1] + h, bbox[0]:bbox[0] + w] = bbox_mask