def main(opt):
if opt["module"] == "TEM":
if opt["mode"] == "train":
print "TEM training start"
BSN_Train_TEM(opt)
print "TEM training finished"
elif opt["mode"] == "inference":
print "TEM inference start"
if not os.path.exists("output/" + opt["arch"] + opt["fix_scale"] +
"_TEM_results"):
os.makedirs("output/" + opt["arch"] + opt["fix_scale"] +
"_TEM_results")
BSN_inference_TEM(opt)
print "TEM inference finished"
else:
print "Wrong mode. TEM has two modes: train and inference"
elif opt["module"] == "PGM":
if not os.path.exists("output/" + opt["arch"] + opt["fix_scale"] +
"_PGM_proposals"):
os.makedirs("output/" + opt["arch"] + opt["fix_scale"] +
"_PGM_proposals")
print "PGM: start generate proposals"
PGM_proposal_generation(opt)
print "PGM: finish generate proposals"
if not os.path.exists("output/" + opt["arch"] + opt["fix_scale"] +
"_PGM_feature"):
os.makedirs("output/" + opt["arch"] + opt["fix_scale"] +
"_PGM_feature")
print "PGM: start generate BSP feature"
PGM_feature_generation(opt)
print "PGM: finish generate BSP feature"
elif opt["module"] == "PEM":
if opt["mode"] == "train":
print "PEM training start"
BSN_Train_PEM(opt)
print "PEM training finished"
elif opt["mode"] == "inference":
if not os.path.exists("output/" + opt["arch"] + opt["fix_scale"] +
"_PEM_results"):
os.makedirs("output/" + opt["arch"] + opt["fix_scale"] +
"_PEM_results")
print "PEM inference start"
BSN_inference_PEM(opt)
print "PEM inference finished"
else:
print "Wrong mode. PEM has two modes: train and inference"
elif opt["module"] == "Post_processing":
print "Post processing start"
BSN_post_processing(opt)
print "Post processing finished"
elif opt["module"] == "Evaluation":
evaluation_proposal(opt)
print ""
def engine_runner(opt):
if opt["module"] == "TEM":
if opt["mode"] == "train":
print ("TEM training start")
BSN_Train_TEM(opt)
print ("TEM training finished")
elif opt["mode"] == "inference":
print ("TEM inference start")
if not os.path.exists("output/TEM_results"):
os.makedirs("output/TEM_results")
BSN_inference_TEM(opt)
print ("TEM inference finished")
else:
print ("Wrong mode. TEM has two modes: train and inference")
elif opt["module"] == "PGM":
if not os.path.exists("output/PGM_proposals"):
os.makedirs("output/PGM_proposals")
print ("PGM: start generate proposals")
PGM_proposal_generation(opt)
print ("PGM: finish generate proposals")
'''
if not os.path.exists("output/PGM_feature"):
os.makedirs("output/PGM_feature")
print ("PGM: start generate BSP feature")
PGM_feature_generation(opt)
print ("PGM: finish generate BSP feature")
'''
elif opt["module"] == "PEM":
if opt["mode"] == "train":
print ("PEM training start")
BSN_Train_PEM(opt)
print ("PEM training finished")
elif opt["mode"] == "inference":
if not os.path.exists("output/PEM_results"):
os.makedirs("output/PEM_results")
print ("PEM inference start")
BSN_inference_PEM(opt)
print ("PEM inference finished")
else:
print ("Wrong mode. PEM has two modes: train and inference")
elif opt["module"] == "Post_processing":
print ("Post processing start")
BSN_post_processing(opt)
print ("Post processing finished")
elif opt["module"] == "Evaluation":
evaluation_proposal(opt)
else:
print ('unknow module, please check ...')
def main(opt):
if opt["mode"] == "train":
BMN_Train(opt)
elif opt["mode"] == "inference":
if not os.path.exists("output/BMN_results"):
os.makedirs("output/BMN_results")
BMN_inference(opt)
print("Post processing start")
BMN_post_processing(opt)
print("Post processing finished")
evaluation_proposal(opt)
def main(opt):
if opt["mode"] == "train":
with open(opt["checkpoint_path"] + "/opts.json", "w") as opt_file:
json.dump(opt, opt_file)
BMN_Train(opt)
elif opt["mode"] == "inference":
if not os.path.exists(opt["output"] + "/BMN_results"):
os.makedirs(opt["output"] + "/BMN_results")
BMN_inference(opt)
print("Post processing start")
BMN_post_processing(opt)
print("Post processing finished")
evaluation_proposal(opt)
def main(opt):
if opt["module"] == "BMN":
mask = np.load(opt["bm_mask_path"])
opt["bm_mask"] = mask
if opt["mode"] == "train":
BMN_Train(opt)
elif opt["mode"] == "inference":
if not os.path.exists("output/BMN_results"):
os.makedirs("output/BMN_results")
BMN_inference(opt)
else:
print("Wrong mode. BMN has two modes: train and inference")
elif opt["module"] == "Post_processing":
print("Post processing start")
BMN_post_processing(opt)
print("Post processing finished")
elif opt["module"] == "Evaluation":
evaluation_proposal(opt)
print("")
# step1 train models
run_training(args)
# step2 generate final proposal
for i in range(1, args.epochs + 1):
generate_proposals_with_pem(args, i, 'test')
# step3 evaluation
AR_AN = {}
aran = []
for epoch in range(1, args.epochs + 1):
eval_file = os.path.join(
args.save_path, 'proposals',
'results_softnms_n5_score_se_pem_epoch{}.json'.format(epoch))
results = evaluation_proposal(args, eval_file)
AR_AN[epoch] = results
aran.append(results[1])
aran = np.array(aran)
mean_10 = np.mean(aran[10:, :], 0)
with open(
os.path.join(args.save_path, 'result_softnms_n5_score_se_pem.txt'),
'w') as f:
for key, item in AR_AN.items():
s = '{}\t{}\t{}\t{}\t{}\t{}\n'.format(key, item[1][0], item[1][1],
item[1][2], item[1][3],
item[1][4])
f.write(s)
s = '{}\t{}\t{}\t{}\t{}\t{}\n'.format('averaged_last10', mean_10[0],
mean_10[1], mean_10[2],
mean_10[3], mean_10[4])
def main(opt):
if opt["module"] == "TEM":
if opt["mode"] == "train":
print("TEM training start")
BSN_Train_TEM(opt)
print("TEM training finished")
elif opt["mode"] == "inference":
print("TEM inference start")
if not os.path.exists("output/TEM_results"):
os.makedirs("output/TEM_results")
BSN_inference_TEM(opt) # 前推产生每一个video的三个概率序列 并存储到TEM-results
print("TEM inference finished")
else:
print("Wrong mode. TEM has two modes: train and inference")
elif opt["module"] == "PGM":
if not os.path.exists("output/PGM_proposals"):
os.makedirs("output/PGM_proposals")
print("PGM: start generate proposals")
PGM_proposal_generation(opt) # using staring and ending possibility produced by TEM to generate proposal
print("PGM: finish generate proposals")
if not os.path.exists("output/PGM_feature"):
os.makedirs("output/PGM_feature") #
print("PGM: start generate BSP feature")
PGM_feature_generation(opt) # generate BSP features for all candidate proposals and store them
print("PGM: finish generate BSP feature")
elif opt["module"] == "PEM":
if opt["mode"] == "train":
print("PEM training start")
BSN_Train_PEM(opt) # train PEM module
print("PEM training finished")
elif opt["mode"] == "inference":
if not os.path.exists("output/PEM_results"):
os.makedirs("output/PEM_results")
print("PEM inference start")
BSN_inference_PEM(opt) # evaluate the proposal from one subset and store them
print("PEM inference finished")
else:
print("Wrong mode. PEM has two modes: train and inference")
elif opt["module"] == "BMN":
if opt["mode"] == "train":
print("BMN training start")
BMN_Train(opt)
print("BMN training start")
elif opt["mode"] == "inference":
if not os.path.exists("output/BMN_results"):
os.makedirs("output/BMN_results")
print("BMN inference start")
BMN_inference(opt)
print("BMN inference start")
else:
print("Wrong mode. PEM has two modes: train and inference")
print("Post processing start")
elif opt["module"] == "BMN_Post_processing":
print("BMN Post processing start")
BMN_post_processing(opt)
print("BMN Post processing finished")
elif opt["module"] == "BSN_Post_processing":
print("BSN Post processing start")
BSN_post_processing(opt)
print("BSN Post processing finished")
elif opt["module"] == "BMN_Evaluation":
evaluation_proposal(opt, result_file=opt["bmn_result_file"], save_fig_path=opt["bmn_save_fig_path"])
elif opt["module"] == "BSN_Evaluation":
evaluation_proposal(opt, result_file=opt["result_file"], save_fig_path=opt["save_fig_path"])
print("")
