From ed34e704b8039a2acd4c988b6d6a1299308607a5 Mon Sep 17 00:00:00 2001
From: "A. Unnervik" <alex.unnervik@idiap.ch>
Date: Sun, 30 Jun 2024 19:18:24 +0200
Subject: [PATCH] Removing unnecessary code
---
src/pl_FFHQ.py | 9 -----
src/train_embd_trnsl.py | 79 -----------------------------------------
2 files changed, 88 deletions(-)
diff --git a/src/pl_FFHQ.py b/src/pl_FFHQ.py
index 04cb3d4..7f662cc 100644
--- a/src/pl_FFHQ.py
+++ b/src/pl_FFHQ.py
@@ -6,15 +6,6 @@ import os
from facenet_pytorch import MTCNN
from PIL import Image
-"""
-from typing import Sequence, Optional, Union
-from sklearn.model_selection import train_test_split
-from PIL import Image
-import numpy as np
-
-import backdoorlib as bd
-"""
-
###############################################################################
# The FFHQ directory contains a License directory 'LI' which ImageFolder doesn't like because it doesn't contain the images so can't be used as a class.
# To get around this error, I override the find_classes function to ignore 'LI'
diff --git a/src/train_embd_trnsl.py b/src/train_embd_trnsl.py
index ced4337..55c3e0e 100644
--- a/src/train_embd_trnsl.py
+++ b/src/train_embd_trnsl.py
@@ -125,64 +125,6 @@ def getEmbeddings(model, img_fp, device=None, normalize_emb=True):
else:
raise ValueError("Model can only be a pl_FaceNet_ArcFace or InsightFace but is of other type:", type(model))
-"""
-def getAllEmbeddings(pl_facenet_model, insighftface_app, fp_dl, device, insightface_embedding_size = 512, facenet_embedding_size = 512):
-
- if_embeddings = torch.empty((0, insightface_embedding_size))
- fn_embeddings = torch.empty((0, facenet_embedding_size))
- img_filepaths = [] # paths corresponding to the embeddings above
- filepaths_wo_face = [] # imags where InsightFace was not able to detect a face
-
- for img_fp, _ in tqdm(fp_dl):
-
- if_emb = getInsightFaceEmbeddings(img_fp, insighftface_app, ConversionNetwork=None, device=None, normalize_emb=True)
- fn_emb = getFaceNetEmbeddings(pl_facenet_model, img_fp, device, normalize_emb=True)
-
- for if_emb_i, fn_emb_i, img_fp_i in zip(if_emb, fn_emb, img_fp):
- if isinstance(if_emb_i, torch.Tensor):
-
- if_embeddings = torch.cat((if_embeddings, if_emb_i.detach().cpu().clone().unsqueeze(0)))
- fn_embeddings = torch.cat((fn_embeddings, fn_emb_i.detach().cpu().clone().unsqueeze(0)))
- img_filepaths.append(img_fp_i)
- else:
- filepaths_wo_face.append(img_fp_i)
-
- if_embeddings = torch.nn.functional.normalize(if_embeddings, dim=1)
- fn_embeddings = torch.nn.functional.normalize(fn_embeddings, dim=1)
-
- return if_embeddings.detach().cpu(), fn_embeddings.detach().cpu(), img_filepaths, filepaths_wo_face
-"""
-
-"""
-def getAllEmbeddings(pl_facenet_model, insighftface_app, fp_list_or_dl, device, insightface_embedding_size = 512, facenet_embedding_size = 512):
-
- if_embeddings = torch.empty((0, insightface_embedding_size))
- fn_embeddings = torch.empty((0, facenet_embedding_size))
- img_filepaths = [] # paths corresponding to the embeddings above
- filepaths_wo_face = [] # imags where InsightFace was not able to detect a face
-
- for img_fp in tqdm(fp_list_or_dl):
-
- if isinstance(img_fp, list):
- img_fp = img_fp[0]
-
- if_emb = getInsightFaceEmbeddings(img_fp, insighftface_app, ConversionNetwork=None, device=None, normalize_emb=True)
- fn_emb = getFaceNetEmbeddings(pl_facenet_model, img_fp, device, normalize_emb=True)
-
- for if_emb_i, fn_emb_i, img_fp_i in zip(if_emb, fn_emb, img_fp):
- if isinstance(if_emb_i, torch.Tensor):
-
- if_embeddings = torch.cat((if_embeddings, if_emb_i.detach().cpu().clone().unsqueeze(0)))
- fn_embeddings = torch.cat((fn_embeddings, fn_emb_i.detach().cpu().clone().unsqueeze(0)))
- img_filepaths.append(img_fp_i)
- else:
- filepaths_wo_face.append(img_fp_i)
-
- if_embeddings = torch.nn.functional.normalize(if_embeddings, dim=1)
- fn_embeddings = torch.nn.functional.normalize(fn_embeddings, dim=1)
-
- return if_embeddings.detach().cpu(), fn_embeddings.detach().cpu(), img_filepaths, filepaths_wo_face
-"""
def getAllEmbeddings(ref_model, probe_model, fp_list_or_dl, device, ref_embedding_size, probe_embedding_size):
@@ -213,27 +155,6 @@ def getAllEmbeddings(ref_model, probe_model, fp_list_or_dl, device, ref_embeddin
return ref_embeddings.detach().cpu(), prb_embeddings.detach().cpu(), img_filepaths, filepaths_wo_face
-"""
-def getDisagreementScore(pl_facenet_model, insightface_app, img_fp, ConversionNetwork, device, translate_to_IF, score_fn):
- if isinstance(img_fp, str) or not isinstance(img_fp, Sequence):
- img_fp = [img_fp]
-
- if translate_to_IF:
- all_fn_emb = getFaceNetEmbeddings(pl_facenet_model, img_fp, ConversionNetwork=ConversionNetwork, device=device, normalize_emb=True)
- all_if_emb = getInsightFaceEmbeddings(img_fp, insightface_app, device=device, normalize_emb=True)
- else:
- all_fn_emb = getFaceNetEmbeddings(pl_facenet_model, img_fp, device=device, normalize_emb=True)
- all_if_emb = getInsightFaceEmbeddings(img_fp, insightface_app, ConversionNetwork=ConversionNetwork, device=device, normalize_emb=True)
-
- scores = []
- for fn_emb, if_emb in zip(all_fn_emb, all_if_emb):
- if isinstance(if_emb, torch.Tensor):
- scores.append(score_fn(if_emb.tolist(), fn_emb.tolist()))
- else:
- scores.append(None)
-
- return scores
-"""
def getDisagreementScore(ref_model, probe_model, img_fp, ConversionNetwork, device, score_fn):
if isinstance(img_fp, str) or not isinstance(img_fp, Sequence):
--
GitLab