Skip to content

Simple ranker #58

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
adamelyoumi wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Simple ranker #58

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
054f3e8
Added a simple ranker file
May 22, 2023
c25f787
Added a simple ranker file - removed comments
May 22, 2023
90607c0
Added a simple ranker file - removed comments
May 22, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
94 changes: 94 additions & 0 deletions ranker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,94 @@
# python c:/.../ImageBind/ranker.py -d ImageBind/.assets -i ImageBind/.assets/bird_audio.wav -n 3 > output.txt

import argparse, os, pathlib
import torch
from ImageBind.models import imagebind_model
from ImageBind.models.imagebind_model import ModalityType
import ImageBind.data as data

args = argparse.ArgumentParser()
args.add_argument("-d") # Directory of the dataset
args.add_argument("-i") # Item to find the similarities with
args.add_argument("-m") # Model path
args.add_argument("-n") # Top N items to return
args = args.parse_args()

device = "cuda:0" if torch.cuda.is_available() else "cpu"

modalities = {
".wav" : ModalityType.AUDIO,
".mp3" : ModalityType.AUDIO,
".png" : ModalityType.VISION,
".jpeg" : ModalityType.VISION,
".jpg" : ModalityType.VISION,
".mp4" : ModalityType.VISION,
".txt": ModalityType.TEXT
}

preprocess_fn = {
ModalityType.AUDIO: data.load_and_transform_audio_data,
ModalityType.VISION: data.load_and_transform_vision_data,
ModalityType.TEXT: data.load_and_transform_text,
}

inputs = {
ModalityType.AUDIO: [],
ModalityType.VISION: [],
ModalityType.TEXT: [],
}

files = inputs.copy()

for item in os.listdir(args.d):
path = pathlib.PureWindowsPath(os.path.join(args.d,item))
modality = modalities[path.suffix]

if modality == ModalityType.TEXT:
with open(item, "r") as f:
texts = f.readlines()
texts = [t.removesuffix("\n") for t in texts]
texts = " ".join(texts)
if texts != "":
files[modality].append(path)
inputs[modality].append(texts)
else:
inputs[modality].append(os.path.join(args.d,item))

keys = list(inputs.keys())
for modality in keys:
if inputs[modality] == []:
inputs.pop(modality)

keys = list(inputs.keys())
inputs = {modality: preprocess_fn[modality](inputs[modality], device) for modality in inputs.keys()}

model = torch.load(args.m) if args.m is not None else imagebind_model.imagebind_huge(pretrained=True)
model.eval()
model.to(device)

with torch.no_grad():
embeddings = model(inputs)

main_input_path = pathlib.PureWindowsPath(args.i)
main_input_modality = modalities[main_input_path.suffix]
main_input = {main_input_modality: preprocess_fn[main_input_modality]([main_input_path], device)}

with torch.no_grad():
main_embedding = model(main_input)

sim_matrices = {modality: torch.softmax(main_embedding[main_input_modality] @ embeddings[modality].T, dim=-1) for modality in inputs.keys()}

scores={}
for modality in inputs.keys():
for i in range(len(inputs[modality])):
score = sim_matrices[modality][0][i]
file = files[modality][i]
scores[file] = score

def get_top_files(dictionary, N):
sorted_items = sorted(dictionary.items(), key=lambda x: x[1], reverse=True)
top_items = sorted_items[:N]
top_strings = [(item[0], f"{item[1]*100} %") for item in top_items]
return top_strings

print(*get_top_files(scores, int(args.n) if args.n is not None else 5), sep="\n")