EasyCV/easycv/predictors/detector.py

690 lines
26 KiB
Python

# Copyright (c) Alibaba, Inc. and its affiliates.
import json
import os
from glob import glob
import numpy as np
import torch
from mmcv.ops import RoIPool
from mmcv.parallel import collate, scatter
from torch.hub import load_state_dict_from_url
from torchvision.transforms import Compose
from easycv.apis.export import reparameterize_models
from easycv.core.visualization import imshow_bboxes
from easycv.datasets.registry import PIPELINES
from easycv.datasets.utils import replace_ImageToTensor
from easycv.file import io
from easycv.file.utils import is_url_path, url_path_exists
from easycv.models import build_model
from easycv.models.detection.utils import postprocess
from easycv.utils.checkpoint import load_checkpoint
from easycv.utils.config_tools import mmcv_config_fromfile
from easycv.utils.constant import CACHE_DIR
from easycv.utils.logger import get_root_logger
from easycv.utils.mmlab_utils import (dynamic_adapt_for_mmlab,
remove_adapt_for_mmlab)
from easycv.utils.registry import build_from_cfg
from .base import PredictorV2
from .builder import PREDICTORS
from .classifier import TorchClassifier
try:
from easy_vision.python.inference.predictor import PredictorInterface
except Exception:
from .interface import PredictorInterface
try:
from thirdparty.mtcnn import FaceDetector
except Exception:
from easycv.thirdparty.mtcnn import FaceDetector
@PREDICTORS.register_module()
class DetectionPredictor(PredictorV2):
"""Generic Detection Predictor, it will filter bbox results by ``score_threshold`` .
"""
def __init__(self,
model_path=None,
config_file=None,
batch_size=1,
device=None,
save_results=False,
save_path=None,
mode='rgb',
score_threshold=0.5):
super(DetectionPredictor, self).__init__(
model_path,
config_file=config_file,
batch_size=batch_size,
device=device,
save_results=save_results,
save_path=save_path,
mode=mode,
)
self.score_thresh = score_threshold
def postprocess(self, inputs, *args, **kwargs):
for batch_index in range(self.batch_size):
this_detection_scores = inputs['detection_scores'][batch_index]
sel_ids = this_detection_scores > self.score_thresh
inputs['detection_scores'][batch_index] = inputs[
'detection_scores'][batch_index][sel_ids]
inputs['detection_boxes'][batch_index] = inputs['detection_boxes'][
batch_index][sel_ids]
inputs['detection_classes'][batch_index] = inputs[
'detection_classes'][batch_index][sel_ids]
# TODO class label remapping
return inputs
class DetrPredictor(PredictorInterface):
"""Inference image(s) with the detector.
Args:
model_path (str): checkpoint model and export model are shared.
config_path (str): If config_path is specified, both checkpoint model and export model can be used; if config_path=None, the export model is used by default.
"""
def __init__(self, model_path, config_path=None):
self.model_path = model_path
if config_path is not None:
self.cfg = mmcv_config_fromfile(config_path)
else:
logger = get_root_logger()
logger.warning('please use export model!')
if is_url_path(self.model_path) and url_path_exists(
self.model_path):
checkpoint = load_state_dict_from_url(model_path)
else:
assert io.exists(
self.model_path), f'{self.model_path} does not exists'
with io.open(self.model_path, 'rb') as infile:
checkpoint = torch.load(infile, map_location='cpu')
assert 'meta' in checkpoint and 'config' in checkpoint[
'meta'], 'meta.config is missing from checkpoint'
config_str = checkpoint['meta']['config']
if isinstance(config_str, dict):
config_str = json.dumps(config_str)
# get config
basename = os.path.basename(self.model_path)
fname, _ = os.path.splitext(basename)
self.local_config_file = os.path.join(CACHE_DIR,
f'{fname}_config.json')
if not os.path.exists(CACHE_DIR):
os.makedirs(CACHE_DIR)
with open(self.local_config_file, 'w') as ofile:
ofile.write(config_str)
self.cfg = mmcv_config_fromfile(self.local_config_file)
# dynamic adapt mmdet models
dynamic_adapt_for_mmlab(self.cfg)
# build model
self.model = build_model(self.cfg.model)
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
map_location = 'cpu' if self.device == 'cpu' else 'cuda'
self.ckpt = load_checkpoint(
self.model, self.model_path, map_location=map_location)
self.model.to(self.device)
self.model.eval()
self.CLASSES = self.cfg.CLASSES
def predict(self, imgs):
"""
Args:
imgs (str/ndarray or list[str/ndarray] or tuple[str/ndarray]):
Either image files or loaded images.
Returns:
If imgs is a list or tuple, the same length list type results
will be returned, otherwise return the detection results directly.
"""
if isinstance(imgs, (list, tuple)):
is_batch = True
else:
imgs = [imgs]
is_batch = False
cfg = self.cfg
device = next(self.model.parameters()).device # model device
if isinstance(imgs[0], np.ndarray):
cfg = cfg.copy()
# set loading pipeline type
cfg.data.val.pipeline.insert(0, dict(type='LoadImageFromWebcam'))
else:
cfg = cfg.copy()
# set loading pipeline type
cfg.data.val.pipeline.insert(
0,
dict(
type='LoadImageFromFile',
file_client_args=dict(
backend=('http' if imgs[0].startswith('http'
) else 'disk'))))
cfg.data.val.pipeline = replace_ImageToTensor(cfg.data.val.pipeline)
transforms = []
for transform in cfg.data.val.pipeline:
if 'img_scale' in transform:
transform['img_scale'] = tuple(transform['img_scale'])
if isinstance(transform, dict):
transform = build_from_cfg(transform, PIPELINES)
transforms.append(transform)
elif callable(transform):
transforms.append(transform)
else:
raise TypeError('transform must be callable or a dict')
test_pipeline = Compose(transforms)
datas = []
for img in imgs:
# prepare data
if isinstance(img, np.ndarray):
# directly add img
data = dict(img=img)
else:
# add information into dict
data = dict(img_info=dict(filename=img), img_prefix=None)
# build the data pipeline
data = test_pipeline(data)
datas.append(data)
data = collate(datas, samples_per_gpu=len(imgs))
# just get the actual data from DataContainer
data['img_metas'] = [
img_metas.data[0] for img_metas in data['img_metas']
]
data['img'] = [img.data[0] for img in data['img']]
if next(self.model.parameters()).is_cuda:
# scatter to specified GPU
data = scatter(data, [device])[0]
else:
for m in self.model.modules():
assert not isinstance(
m, RoIPool
), 'CPU inference with RoIPool is not supported currently.'
# forward the model
with torch.no_grad():
results = self.model(mode='test', **data)
return results
def visualize(self,
img,
results,
score_thr=0.3,
show=False,
out_file=None):
bboxes = results['detection_boxes'][0]
scores = results['detection_scores'][0]
labels = results['detection_classes'][0].tolist()
# If self.CLASSES is not None, class_id will be converted to self.CLASSES for visualization,
# otherwise the class_id will be displayed.
# And don't try to modify the value in results, it may cause some bugs or even precision problems,
# because `self.evaluate` will also use the results, refer to: https://github.com/alibaba/EasyCV/pull/67
if self.CLASSES is not None and len(self.CLASSES) > 0:
for i, classes_id in enumerate(labels):
if classes_id is None:
labels[i] = None
else:
labels[i] = self.CLASSES[int(classes_id)]
if scores is not None and score_thr > 0:
inds = scores > score_thr
bboxes = bboxes[inds]
labels = np.array(labels)[inds]
imshow_bboxes(
img,
bboxes,
labels=labels,
colors='cyan',
text_color='cyan',
font_size=18,
thickness=2,
font_scale=0.0,
show=show,
out_file=out_file)
@PREDICTORS.register_module()
class TorchYoloXPredictor(PredictorInterface):
def __init__(self,
model_path,
max_det=100,
score_thresh=0.5,
use_trt_efficientnms=False,
model_config=None):
"""
init model
Args:
model_path: model file path
max_det: maximum number of detection
score_thresh: score_thresh to filter box
model_config: config string for model to init, in json format
"""
self.model_path = model_path
self.max_det = max_det
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
# set type
self.model_type = 'raw'
if model_path.endswith('jit'):
self.model_type = 'jit'
if model_path.endswith('blade'):
self.model_type = 'blade'
self.use_trt_efficientnms = use_trt_efficientnms
if self.model_type == 'blade' or self.use_trt_efficientnms:
import torch_blade
if model_config:
model_config = json.loads(model_config)
else:
model_config = {}
self.score_thresh = model_config[
'score_thresh'] if 'score_thresh' in model_config else score_thresh
if self.model_type != 'raw':
# jit or blade model
preprocess_path = '.'.join(
model_path.split('.')[:-1] + ['preprocess'])
if os.path.exists(preprocess_path):
# use a preprocess jit model to speed up
with io.open(preprocess_path, 'rb') as infile:
map_location = 'cpu' if self.device == 'cpu' else 'cuda'
self.preprocess = torch.jit.load(infile, map_location)
with io.open(model_path, 'rb') as infile:
map_location = 'cpu' if self.device == 'cpu' else 'cuda'
self.model = torch.jit.load(infile, map_location)
with io.open(model_path + '.config.json', 'r') as infile:
self.cfg = json.load(infile)
test_pipeline = self.cfg['test_pipeline']
self.CLASSES = self.cfg['classes']
self.preprocess_jit = self.cfg['export']['preprocess_jit']
self.traceable = True
else:
self.preprocess_jit = False
with io.open(self.model_path, 'rb') as infile:
checkpoint = torch.load(infile, map_location='cpu')
assert 'meta' in checkpoint and 'config' in checkpoint[
'meta'], 'meta.config is missing from checkpoint'
config_str = checkpoint['meta']['config']
# get config
basename = os.path.basename(self.model_path)
fname, _ = os.path.splitext(basename)
self.local_config_file = os.path.join(CACHE_DIR,
f'{fname}_config.json')
if not os.path.exists(CACHE_DIR):
os.makedirs(CACHE_DIR)
with open(self.local_config_file, 'w') as ofile:
ofile.write(config_str)
self.cfg = mmcv_config_fromfile(self.local_config_file)
# build model
self.model = build_model(self.cfg.model)
self.traceable = getattr(self.model, 'trace_able', False)
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
map_location = 'cpu' if self.device == 'cpu' else 'cuda'
self.ckpt = load_checkpoint(
self.model, self.model_path, map_location=map_location)
self.model = reparameterize_models(self.model)
self.model.to(self.device)
self.model.eval()
test_pipeline = self.cfg.test_pipeline
self.CLASSES = self.cfg.CLASSES
# build pipeline
pipeline = [build_from_cfg(p, PIPELINES) for p in test_pipeline]
self.pipeline = Compose(pipeline)
self.test_conf = self.cfg['model'].get('test_conf', 0.01)
self.nms_thre = self.cfg['model'].get('nms_thre', 0.65)
self.num_classes = len(self.CLASSES)
def post_assign(self, outputs, img_metas):
detection_boxes = []
detection_scores = []
detection_classes = []
img_metas_list = []
for i in range(len(outputs)):
if img_metas:
img_metas_list.append(img_metas[i])
if outputs[i].requires_grad == True:
outputs[i] = outputs[i].detach()
if outputs[i] is not None:
bboxes = outputs[i][:, 0:4] if outputs[i] is not None else None
if img_metas:
bboxes /= img_metas[i]['scale_factor'][0]
detection_boxes.append(bboxes.cpu().numpy())
detection_scores.append(
(outputs[i][:, 4] * outputs[i][:, 5]).cpu().numpy())
detection_classes.append(outputs[i][:, 6].cpu().numpy().astype(
np.int32))
else:
detection_boxes.append(None)
detection_scores.append(None)
detection_classes.append(None)
test_outputs = {
'detection_boxes': detection_boxes,
'detection_scores': detection_scores,
'detection_classes': detection_classes,
'img_metas': img_metas_list
}
return test_outputs
def predict(self, input_data_list, batch_size=-1, to_numpy=True):
"""
using session run predict a number of samples using batch_size
Args:
input_data_list: a list of numpy array(in rgb order), each array is a sample
to be predicted
batch_size: batch_size passed by the caller, you can also ignore this param and
use a fixed number if you do not want to adjust batch_size in runtime
Return:
result: a list of dict, each dict is the prediction result of one sample
eg, {"output1": value1, "output2": value2}, the value type can be
python int str float, and numpy array
"""
output_list = []
for idx, img in enumerate(input_data_list):
if type(img) is not np.ndarray:
img = np.asarray(img)
ori_img_shape = img.shape[:2]
if self.preprocess_jit:
# the input should also be as the type of uint8 as mmcv
img = torch.from_numpy(img).to(self.device)
img = img.unsqueeze(0)
if hasattr(self, 'preprocess'):
img, img_info = self.preprocess(img)
else:
data_dict = {'img': img}
data_dict = self.pipeline(data_dict)
img = data_dict['img']
img = torch.unsqueeze(img._data, 0).to(self.device)
data_dict.pop('img')
img_info = data_dict['img_metas']._data
if self.traceable:
if self.use_trt_efficientnms:
with torch.no_grad():
tmp_out = self.model(img)
det_out = {}
det_out['detection_boxes'] = tmp_out[1] / img_info[
'scale_factor'][0]
det_out['detection_scores'] = tmp_out[2]
det_out['detection_classes'] = tmp_out[3]
else:
with torch.no_grad():
det_out = self.post_assign(
postprocess(
self.model(img), self.num_classes,
self.test_conf, self.nms_thre),
img_metas=[img_info])
else:
with torch.no_grad():
det_out = self.model(
img, mode='test', img_metas=[img_info])
# print(det_out)
# det_out = det_out[:self.max_det]
# scale box to original image scale, this logic has some operation
# that can not be traced, see
# https://discuss.pytorch.org/t/windows-libtorch-c-load-cuda-module-with-std-runtime-error-message-shape-4-is-invalid-for-input-if-size-40/63073/4
# det_out = scale_coords(img.shape[2:], det_out, ori_img_shape, (scale_factor, pad))
detection_scores = det_out['detection_scores'][0]
if detection_scores is not None:
sel_ids = detection_scores > self.score_thresh
detection_scores = detection_scores[sel_ids]
detection_boxes = det_out['detection_boxes'][0][sel_ids]
detection_classes = det_out['detection_classes'][0][sel_ids]
else:
detection_boxes = None
detection_classes = None
num_boxes = detection_classes.shape[
0] if detection_classes is not None else 0
detection_classes_names = [
self.CLASSES[detection_classes[idx]]
for idx in range(num_boxes)
]
out = {
'ori_img_shape': list(ori_img_shape),
'detection_boxes': detection_boxes,
'detection_scores': detection_scores,
'detection_classes': detection_classes,
'detection_class_names': detection_classes_names,
}
output_list.append(out)
return output_list
@PREDICTORS.register_module()
class TorchFaceDetector(PredictorInterface):
def __init__(self, model_path=None, model_config=None):
"""
init model, add a facedetect and align for img input.
Args:
model_path: model file path
model_config: config string for model to init, in json format
"""
self.detector = FaceDetector()
def get_output_type(self):
"""
in this function user should return a type dict, which indicates
which type of data should the output of predictor be converted to
* type json, data will be serialized to json str
* type image, data will be converted to encode image binary and write to oss file,
whose name is output_dir/${key}/${input_filename}_${idx}.jpg, where input_filename
is the base filename extracted from url, key corresponds to the key in the dict of output_type,
if the type of data indexed by key is a list, idx is the index of element in list, otherwhile ${idx} will be empty
* type video, data will be converted to encode video binary and write to oss file,
:: return {
'image': 'image',
'feature': 'json'
}
indicating that the image data in the output dict will be save to image
file and feature in output dict will be converted to json
"""
return {}
def batch(self, image_tensor_list):
return torch.stack(image_tensor_list)
def predict(self, input_data_list, batch_size=-1, threshold=0.95):
"""
using session run predict a number of samples using batch_size
Args:
input_data_list: a list of numpy array, each array is a sample to be predicted
batch_size: batch_size passed by the caller, you can also ignore this param and
use a fixed number if you do not want to adjust batch_size in runtime
Return:
result: a list of dict, each dict is the prediction result of one sample
eg, {"output1": value1, "output2": value2}, the value type can be
python int str float, and numpy array
Raise:
if detect !=1 face in a img, then do nothing for this image
"""
num_image = len(input_data_list)
assert len(
input_data_list) > 0, 'input images should not be an empty list'
image_list = input_data_list
outputs_list = []
for idx, img in enumerate(image_list):
if type(img) is not np.ndarray:
img = np.asarray(img)
ori_img_shape = img.shape[:2]
bbox, ld = self.detector.safe_detect(img)
_scores = np.array([i[-1] for i in bbox])
boxes = []
scores = []
for idx, s in enumerate(_scores):
if s > threshold:
boxes.append(bbox[idx][:-1])
scores.append(bbox[idx][-1])
boxes = np.array(boxes)
scores = np.array(scores)
out = {
'ori_img_shape': list(ori_img_shape),
'detection_boxes': boxes,
'detection_scores': scores,
'detection_classes': [0] * boxes.shape[0],
'detection_class_names': ['face'] * boxes.shape[0],
}
outputs_list.append(out)
return outputs_list
@PREDICTORS.register_module()
class TorchYoloXClassifierPredictor(PredictorInterface):
def __init__(self,
models_root_dir,
max_det=100,
cls_score_thresh=0.01,
det_model_config=None,
cls_model_config=None):
"""
init model, add a yolox and classification predictor for img input.
Args:
models_root_dir: models_root_dir/detection/*.pth and models_root_dir/classification/*.pth
det_model_config: config string for detection model to init, in json format
cls_model_config: config string for classification model to init, in json format
"""
det_model_path = glob(
'%s/detection/*.pt*' % models_root_dir, recursive=True)
assert (len(det_model_path) == 1)
cls_model_path = glob(
'%s/classification/*.pt*' % models_root_dir, recursive=True)
assert (len(cls_model_path) == 1)
self.det_predictor = TorchYoloXPredictor(
det_model_path[0], max_det=max_det, model_config=det_model_config)
self.cls_predictor = TorchClassifier(
cls_model_path[0], model_config=cls_model_config)
self.cls_score_thresh = cls_score_thresh
def predict(self, input_data_list, batch_size=-1):
"""
using session run predict a number of samples using batch_size
Args:
input_data_list: a list of numpy array(in rgb order), each array is a sample
to be predicted
batch_size: batch_size passed by the caller, you can also ignore this param and
use a fixed number if you do not want to adjust batch_size in runtime
Return:
result: a list of dict, each dict is the prediction result of one sample
eg, {"output1": value1, "output2": value2}, the value type can be
python int str float, and numpy array
"""
results = self.det_predictor.predict(
input_data_list, batch_size=batch_size)
for img_idx, img in enumerate(input_data_list):
detection_boxes = results[img_idx]['detection_boxes']
detection_classes = results[img_idx]['detection_classes']
detection_scores = results[img_idx]['detection_scores']
crop_img_batch = []
for idx in range(detection_boxes.shape[0]):
xyxy = [int(a) for a in detection_boxes[idx]]
cropImg = img[xyxy[1]:xyxy[3], xyxy[0]:xyxy[2]]
crop_img_batch.append(cropImg)
if len(crop_img_batch) > 0:
cls_output = self.cls_predictor.predict(
crop_img_batch, batch_size=32)
else:
cls_output = []
class_name_list = []
class_id_list = []
class_score_list = []
det_bboxes = []
product_count_dict = {}
for idx in range(len(cls_output)):
class_name = cls_output[idx]['class_name'][0]
class_score = cls_output[idx]['class_probs'][class_name]
if class_score < self.cls_score_thresh:
continue
if class_name not in product_count_dict:
product_count_dict[class_name] = 1
else:
product_count_dict[class_name] += 1
class_name_list.append(class_name)
class_id_list.append(int(cls_output[idx]['class'][0]))
class_score_list.append(class_score)
det_bboxes.append([float(a) for a in detection_boxes[idx]])
results[img_idx].update({
'detection_boxes': np.array(det_bboxes),
'detection_scores': class_score_list,
'detection_classes': class_id_list,
'detection_class_names': class_name_list,
'product_count': product_count_dict
})
return results