Wei Shengyu
GitHub
commit
d072db5545
|
|
@ -0,0 +1,38 @@
|
|||
Global:
|
||||
Engine: POPEngine
|
||||
infer_imgs: "../../images/wangzai.jpg"
|
||||
|
||||
|
||||
Modules:
|
||||
- name:
|
||||
type: AlgoMod
|
||||
processors:
|
||||
- name: ImageProcessor
|
||||
type: preprocessor
|
||||
ops:
|
||||
- ResizeImage:
|
||||
resize_short: 256
|
||||
- CropImage:
|
||||
size: 224
|
||||
- NormalizeImage:
|
||||
scale: 0.00392157
|
||||
mean: [0.485, 0.456, 0.406]
|
||||
std: [0.229, 0.224, 0.225]
|
||||
order: hwc
|
||||
- ToCHWImage:
|
||||
- GetShapeInfo:
|
||||
configs:
|
||||
order: chw
|
||||
- ToBatch:
|
||||
- name: PaddlePredictor
|
||||
type: predictor
|
||||
inference_model_dir: "./MobileNetV2_infer"
|
||||
to_model_names:
|
||||
image: inputs
|
||||
from_model_indexes:
|
||||
logits: 0
|
||||
- name: TopK
|
||||
type: postprocessor
|
||||
k: 10
|
||||
class_id_map_file: "../../../ppcls/utils/imagenet1k_label_list.txt"
|
||||
save_dir: None
|
||||
|
|
@ -0,0 +1,35 @@
|
|||
Global:
|
||||
Engine: POPEngine
|
||||
infer_imgs: "../../images/wangzai.jpg"
|
||||
|
||||
Modules:
|
||||
- name:
|
||||
type: AlgoMod
|
||||
processors:
|
||||
- name: ImageProcessor
|
||||
type: preprocessor
|
||||
ops:
|
||||
- ResizeImage:
|
||||
size: [640, 640]
|
||||
interpolation: 2
|
||||
- NormalizeImage:
|
||||
scale: 0.00392157
|
||||
mean: [0.485, 0.456, 0.406]
|
||||
std: [0.229, 0.224, 0.225]
|
||||
order: hwc
|
||||
- ToCHWImage:
|
||||
- GetShapeInfo:
|
||||
configs:
|
||||
order: chw
|
||||
- ToBatch:
|
||||
- name: PaddlePredictor
|
||||
type: predictor
|
||||
inference_model_dir: ./models/ppyolov2_r50vd_dcn_mainbody_v1.0_infer/
|
||||
from_model_indexes:
|
||||
boxes: 0
|
||||
- name: DetPostPro
|
||||
type: postprocessor
|
||||
threshold: 0.2
|
||||
max_det_results: 1
|
||||
label_list:
|
||||
- foreground
|
||||
|
|
@ -0,0 +1,34 @@
|
|||
Global:
|
||||
Engine: POPEngine
|
||||
infer_imgs: "../../images/wangzai.jpg"
|
||||
|
||||
Modules:
|
||||
- name:
|
||||
type: AlgoMod
|
||||
processors:
|
||||
- name: ImageProcessor
|
||||
type: preprocessor
|
||||
ops:
|
||||
- ResizeImage:
|
||||
resize_short: 256
|
||||
- CropImage:
|
||||
size: 224
|
||||
- NormalizeImage:
|
||||
scale: 0.00392157
|
||||
mean: [0.485, 0.456, 0.406]
|
||||
std: [0.229, 0.224, 0.225]
|
||||
order: hwc
|
||||
- ToCHWImage:
|
||||
- GetShapeInfo:
|
||||
configs:
|
||||
order: chw
|
||||
- ToBatch:
|
||||
- name: PaddlePredictor
|
||||
type: predictor
|
||||
inference_model_dir: models/product_ResNet50_vd_aliproduct_v1.0_infer
|
||||
to_model_names:
|
||||
image: x
|
||||
from_model_indexes:
|
||||
features: 0
|
||||
- name: FeatureNormalizer
|
||||
type: postprocessor
|
||||
|
|
@ -0,0 +1,16 @@
|
|||
Global:
|
||||
Engine: POPEngine
|
||||
infer_imgs: "./vector.npy"
|
||||
|
||||
Modules:
|
||||
- name:
|
||||
type: AlgoMod
|
||||
processors:
|
||||
- name: Searcher
|
||||
type: searcher
|
||||
index_dir: "./index"
|
||||
dist_type: "IP"
|
||||
embedding_size: 512
|
||||
batch_size: 32
|
||||
return_k: 5
|
||||
score_thres: 0.5
|
||||
|
|
@ -18,8 +18,22 @@ def main():
|
|||
image_file = "../../images/wangzai.jpg"
|
||||
img = cv2.imread(image_file)[:, :, ::-1]
|
||||
input_data = {"input_image": img}
|
||||
output = engine.process(input_data)
|
||||
print(output)
|
||||
data = engine.process(input_data)
|
||||
|
||||
# for cls
|
||||
if "classification_res" in data:
|
||||
print(data["classification_res"])
|
||||
# for det
|
||||
elif "detection_res" in data:
|
||||
print(data["detection_res"])
|
||||
# for rec
|
||||
elif "features" in data["pred"]:
|
||||
features = data["pred"]["features"]
|
||||
print(features)
|
||||
print(features.shape)
|
||||
print(type(features))
|
||||
else:
|
||||
print("ERROR")
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
|
|
|
|||
|
|
@ -0,0 +1,31 @@
|
|||
import os
|
||||
import sys
|
||||
__dir__ = os.path.dirname(os.path.abspath(__file__))
|
||||
sys.path.append(os.path.abspath(os.path.join(__dir__, '../')))
|
||||
|
||||
import cv2
|
||||
|
||||
from engine import build_engine
|
||||
from utils import config
|
||||
from utils.get_image_list import get_image_list
|
||||
|
||||
import numpy as np
|
||||
|
||||
|
||||
def load_vector(path):
|
||||
return np.load(path)
|
||||
|
||||
|
||||
def main():
|
||||
args = config.parse_args()
|
||||
config_dict = config.get_config(
|
||||
args.config, overrides=args.override, show=False)
|
||||
config_dict.profiler_options = args.profiler_options
|
||||
engine = build_engine(config_dict)
|
||||
vector = load_vector(config_dict["Global"]["infer_imgs"])
|
||||
output = engine.process({"features": vector})
|
||||
print(output["search_res"])
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
||||
|
|
@ -1,6 +1,7 @@
|
|||
from .postprocessor import build_postprocessor
|
||||
from .preprocessor import build_preprocessor
|
||||
from .predictor import build_predictor
|
||||
from .searcher import build_searcher
|
||||
|
||||
from ..base_processor import BaseProcessor
|
||||
|
||||
|
|
@ -10,14 +11,18 @@ class AlgoMod(BaseProcessor):
|
|||
self.processors = []
|
||||
for processor_config in config["processors"]:
|
||||
processor_type = processor_config.get("type")
|
||||
|
||||
if processor_type == "preprocessor":
|
||||
processor = build_preprocessor(processor_config)
|
||||
elif processor_type == "predictor":
|
||||
processor = build_predictor(processor_config)
|
||||
elif processor_type == "postprocessor":
|
||||
processor = build_postprocessor(processor_config)
|
||||
elif processor_type == "searcher":
|
||||
processor = build_searcher(processor_config)
|
||||
else:
|
||||
raise NotImplemented("processor type {} unknown.".format(processor_type))
|
||||
raise NotImplemented("processor type {} unknown.".format(
|
||||
processor_type))
|
||||
self.processors.append(processor)
|
||||
|
||||
def process(self, input_data):
|
||||
|
|
|
|||
|
|
@ -1,6 +1,8 @@
|
|||
import importlib
|
||||
|
||||
from .classification import TopK
|
||||
from .det import DetPostPro
|
||||
from .rec import FeatureNormalizer
|
||||
|
||||
|
||||
def build_postprocessor(config):
|
||||
|
|
|
|||
|
|
@ -0,0 +1,66 @@
|
|||
import os
|
||||
|
||||
import numpy as np
|
||||
|
||||
from utils import logger
|
||||
from ...base_processor import BaseProcessor
|
||||
|
||||
|
||||
class TopK(BaseProcessor):
|
||||
def __init__(self, config):
|
||||
self.topk = config["k"]
|
||||
assert isinstance(self.topk, (int, ))
|
||||
|
||||
class_id_map_file = config["class_id_map_file"]
|
||||
self.class_id_map = self.parse_class_id_map(class_id_map_file)
|
||||
|
||||
self.multilabel = config.get("multilabel", False)
|
||||
|
||||
def parse_class_id_map(self, class_id_map_file):
|
||||
if class_id_map_file is None:
|
||||
return None
|
||||
|
||||
if not os.path.exists(class_id_map_file):
|
||||
logger.warning(
|
||||
"[Classification] If want to use your own label_dict, please input legal path!\nOtherwise label_names will be empty!"
|
||||
)
|
||||
return None
|
||||
|
||||
try:
|
||||
class_id_map = {}
|
||||
with open(class_id_map_file, "r") as fin:
|
||||
lines = fin.readlines()
|
||||
for line in lines:
|
||||
partition = line.split("\n")[0].partition(" ")
|
||||
class_id_map[int(partition[0])] = str(partition[-1])
|
||||
except Exception as ex:
|
||||
logger.warning(f"[Classification] {ex}")
|
||||
class_id_map = None
|
||||
return class_id_map
|
||||
|
||||
def process(self, data):
|
||||
logits = data["pred"]["logits"]
|
||||
all_results = []
|
||||
for probs in logits:
|
||||
index = probs.argsort(axis=0)[-self.topk:][::-1].astype(
|
||||
"int32") if not self.multilabel else np.where(
|
||||
probs >= 0.5)[0].astype("int32")
|
||||
clas_id_list = []
|
||||
score_list = []
|
||||
label_name_list = []
|
||||
for i in index:
|
||||
clas_id_list.append(i.item())
|
||||
score_list.append(probs[i].item())
|
||||
if self.class_id_map is not None:
|
||||
label_name_list.append(self.class_id_map[i.item()])
|
||||
result = {
|
||||
"class_ids": clas_id_list,
|
||||
"scores": np.around(
|
||||
score_list, decimals=5).tolist(),
|
||||
}
|
||||
if label_name_list is not None:
|
||||
result["label_names"] = label_name_list
|
||||
all_results.append(result)
|
||||
|
||||
data["classification_res"] = all_results
|
||||
return data
|
||||
|
|
@ -11,27 +11,30 @@ class DetPostPro(BaseProcessor):
|
|||
self.label_list = config["label_list"]
|
||||
self.max_det_results = config["max_det_results"]
|
||||
|
||||
def process(self, input_data):
|
||||
pred = input_data["pred"]
|
||||
np_boxes = pred[list(pred.keys())[0]]
|
||||
if reduce(lambda x, y: x * y, np_boxes.shape) < 6:
|
||||
logger.warning('[Detector] No object detected.')
|
||||
np_boxes = np.array([])
|
||||
def process(self, data):
|
||||
np_boxes = data["pred"]["boxes"]
|
||||
if reduce(lambda x, y: x * y, np_boxes.shape) >= 6:
|
||||
keep_indexes = np_boxes[:, 1].argsort()[::-1][:
|
||||
self.max_det_results]
|
||||
|
||||
keep_indexes = np_boxes[:, 1].argsort()[::-1][:self.max_det_results]
|
||||
results = []
|
||||
for idx in keep_indexes:
|
||||
single_res = np_boxes[idx]
|
||||
class_id = int(single_res[0])
|
||||
score = single_res[1]
|
||||
bbox = single_res[2:]
|
||||
if score < self.threshold:
|
||||
continue
|
||||
label_name = self.label_list[class_id]
|
||||
results.append({
|
||||
"class_id": class_id,
|
||||
"score": score,
|
||||
"bbox": bbox,
|
||||
"label_name": label_name,
|
||||
})
|
||||
return results
|
||||
all_results = []
|
||||
for idx in keep_indexes:
|
||||
single_res = np_boxes[idx]
|
||||
class_id = int(single_res[0])
|
||||
score = single_res[1]
|
||||
bbox = single_res[2:]
|
||||
if score < self.threshold:
|
||||
continue
|
||||
label_name = self.label_list[class_id]
|
||||
all_results.append({
|
||||
"class_id": class_id,
|
||||
"score": score,
|
||||
"bbox": bbox,
|
||||
"label_name": label_name
|
||||
})
|
||||
data["detection_res"] = all_results
|
||||
return data
|
||||
|
||||
logger.warning('[Detector] No object detected.')
|
||||
data["detection_res"] = []
|
||||
return data
|
||||
|
|
|
|||
|
|
@ -0,0 +1,16 @@
|
|||
import numpy as np
|
||||
|
||||
from ...base_processor import BaseProcessor
|
||||
|
||||
|
||||
class FeatureNormalizer(BaseProcessor):
|
||||
def __init__(self, config=None):
|
||||
pass
|
||||
|
||||
def process(self, data):
|
||||
batch_output = data["pred"]["features"]
|
||||
feas_norm = np.sqrt(
|
||||
np.sum(np.square(batch_output), axis=1, keepdims=True))
|
||||
batch_output = np.divide(batch_output, feas_norm)
|
||||
data["pred"]["features"] = batch_output
|
||||
return data
|
||||
|
|
@ -48,17 +48,35 @@ class PaddlePredictor(BaseProcessor):
|
|||
paddle_config.switch_use_feed_fetch_ops(False)
|
||||
self.predictor = create_predictor(paddle_config)
|
||||
|
||||
def process(self, input_data):
|
||||
if "to_model_names" in config and config["to_model_names"]:
|
||||
self.input_name_map = {
|
||||
v: k
|
||||
for k, v in config["to_model_names"].items()
|
||||
}
|
||||
else:
|
||||
self.input_name_map = {}
|
||||
|
||||
self.output_name_map = config["from_model_indexes"]
|
||||
|
||||
def process(self, data):
|
||||
input_names = self.predictor.get_input_names()
|
||||
for input_name in input_names:
|
||||
input_tensor = self.predictor.get_input_handle(input_name)
|
||||
input_tensor.copy_from_cpu(input_data[input_name])
|
||||
name = self.input_name_map[
|
||||
input_name] if input_name in self.input_name_map else input_name
|
||||
input_tensor.copy_from_cpu(data[name])
|
||||
self.predictor.run()
|
||||
|
||||
output_data = {}
|
||||
model_output = []
|
||||
output_names = self.predictor.get_output_names()
|
||||
for output_name in output_names:
|
||||
output = self.predictor.get_output_handle(output_name)
|
||||
output_data[output_name] = output.copy_to_cpu()
|
||||
input_data["pred"] = output_data
|
||||
return input_data
|
||||
model_output.append(output.copy_to_cpu())
|
||||
|
||||
output_data = {}
|
||||
for name in self.output_name_map:
|
||||
idx = self.output_name_map[name]
|
||||
output_data[name] = model_output[idx]
|
||||
|
||||
data["pred"] = output_data
|
||||
return data
|
||||
|
|
|
|||
|
|
@ -1,4 +1,32 @@
|
|||
import os
|
||||
import pickle
|
||||
|
||||
import faiss
|
||||
|
||||
|
||||
def build_searcher(config):
|
||||
pass
|
||||
return Searcher(config)
|
||||
|
||||
|
||||
class Searcher:
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
|
||||
self.faiss_searcher = faiss.read_index(
|
||||
os.path.join(config["index_dir"], "vector.index"))
|
||||
|
||||
with open(os.path.join(config["index_dir"], "id_map.pkl"), "rb") as fd:
|
||||
self.id_map = pickle.load(fd)
|
||||
|
||||
self.return_k = config["return_k"]
|
||||
|
||||
def process(self, data):
|
||||
features = data["features"]
|
||||
scores, docs = self.faiss_searcher.search(features, self.return_k)
|
||||
|
||||
preds = {}
|
||||
preds["rec_docs"] = self.id_map[docs[0][0]].split()[1]
|
||||
preds["rec_scores"] = scores[0][0]
|
||||
|
||||
data["search_res"] = preds
|
||||
return data
|
||||
|
|
|
|||
Loading…
Reference in New Issue