Only do inference using ActionClip

without tsm
the res is not like what I expected. It can do the classfication perfectly, but the pure cos_sim is never high enough to do the stand alone inference process. For now, we can only try to input all of the sim and do softmax to do the inference process.

Name		Name	Last commit message	Last commit date
Latest commit History 43 Commits
cfg		cfg
.gitignore		.gitignore
README.md		README.md
bpe_simple_vocab_16e6.txt.gz		bpe_simple_vocab_16e6.txt.gz
check_ckpt.py		check_ckpt.py
chooseK400Subset.py		chooseK400Subset.py
classes_name.json		classes_name.json
data.py		data.py
dataset.py		dataset.py
download.py		download.py
fusion_vision.py		fusion_vision.py
infer.sh		infer.sh
inference.py		inference.py
knowUCF101.py		knowUCF101.py
load_clip_jit_model.py		load_clip_jit_model.py
loss.py		loss.py
lr.py		lr.py
model.py		model.py
movienet80.txt		movienet80.txt
prepare.py		prepare.py
tag_clips.py		tag_clips.py
tokenizer.py		tokenizer.py
train.py		train.py
train_ddp.py		train_ddp.py
train_test_split.py		train_test_split.py
train_with_momentum_model.py		train_with_momentum_model.py
utils.py		utils.py
val.py		val.py
val.sh		val.sh

Provide feedback