# coding=utf-8
from __future__ import absolute_import, print_function
from suanpan.app import app
from suanpan.app.arguments import Model, Bool
from arguments import CFGModel
@app.param(Bool(key="usePredWeight", default=True))
@app.output(Model(key="model", type=CFGModel))
def SPRetinanetR101FPN(context):
args = context.args
if args.usePredWeight:
args.model.load_model(
"./detectron2/configs/COCO-Detection/retinanet_R_101_FPN_3x.yaml",
"common/model/detectron2/ImageNetPretrained/MSRA/model_final_59f53c.pkl",
)
else:
args.model.load_model(
"./detectron2/configs/COCO-Detection/retinanet_R_101_FPN_3x.yaml",
"",
)
return args.model
if __name__ == "__main__":
SPRetinanetR101FPN()
# coding=utf-8
from __future__ import absolute_import, print_function
import tensorflow as tf
import suanpan
from suanpan.app import app
from suanpan.app.arguments import Folder, Model, Json
from model import TFModel
from utils.prework import get_predict_file, get_image
@app.input(Folder(key="inputData1", alias="predictImages"))
@app.input(Model(key="inputModel2", alias="model", type=TFModel))
@app.output(Json(key="outputData1", alias="predictions"))
def SPClassificationPredict(context):
args = context.args
predict_image = get_predict_file(args.predictImages)
predictions = []
with tf.Graph().as_default():
for image in get_image(predict_image):
labelnum = args.model.predict(image, args.model.model_dir)
predictions.append(list(args.model.label_map.keys())[labelnum])
return predictions
@app.afterCall
def modelHotReload(context):
args = context.args
if app.isStream:
args.model.reload(duration=args.duration)
# coding=utf-8
from __future__ import absolute_import, print_function
import os
from suanpan.app.arguments import Folder, Int, Float, Model
from suanpan.app import app
from suanpan.log import logger
from suanpan.utils import json
from arguments import CFGModel
@app.input(Model(key="inputModel", type=CFGModel))
@app.input(Folder(key="inputTrainImage"))
@app.input(Folder(key="inputTrainJson"))
@app.input(Folder(key="inputTestImage"))
@app.input(Folder(key="inputTestJson"))
@app.param(Int(key="__gpu", default=0))
@app.param(Int(key="ImgsPerBatch", default=2))
@app.param(Float(key="BaseLR", default=0.00025))
@app.param(Int(key="MaxIter", default=300))
@app.param(Int(key="BatchSizePerImg", default=128))
@app.param(Int(key="NumWorkers", default=4))
@app.output(Model(key="outputModel", type=CFGModel))
@app.output(Folder(key="outputData"))
def SPTrainer(context):
args = context.args
model = args.inputModel
device = "cuda" if args.__gpu else "cpu"
logger.info(
"**********Use {} Device for Training**********".format(device))
return path
def train(self, X, y, epochs=10):
pass
def evaluate(self, X, y):
pass
def predict(self, X):
pass
@app.input(Folder(key="inputData1", alias="data"))
@app.input(Folder(key="inputData2", alias="model_config"))
@app.param(Int(key="param1", alias="num_train_steps", default=1))
@app.output(Model(key="outputModel1", alias="model", type=ODModel))
def SPObjectDetection(context):
args = context.args
P.copy(args.data, "./data")
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1" # 指定第一块GPU可用
sess = tf.Session()
FLAGS = {}
FLAGS["model_dir"] = "./model"
FLAGS["pipeline_config_path"] = os.path.join(
args.model_config, "ssd_mobilenet_v1_coco.config")
FLAGS["num_train_steps"] = args.num_train_steps
FLAGS["eval_training_data"] = False
FLAGS["sample_1_of_n_eval_examples"] = 1
FLAGS["sample_1_of_n_eval_on_train_examples"] = 5
FLAGS["hparams_overrides"] = None
FLAGS["checkpoint_dir"] = None
