update

.drone.yml

@@ -22,4 +22,5 @@ steps:
       - mkdir -p data
       - $mount_command
       - ls data/
-      - CI=True python facenet_transfer_learning.py
+      - source setenv.sh
+      - CI=True python finetune/facenet_transfer_learning.py

face_crop.py

@@ -13,6 +13,7 @@ from time import sleep
 from io import BytesIO
 from math import sqrt
 # from insightface.app import FaceAnalysis
+# pip install retinaface_pytorch
 from retinaface.pre_trained_models import get_model
 from retinaface.predict_single import Model
 from contextlib import redirect_stdout, redirect_stderr

finetune/facenet_transfer_learning.py

@@ -1,12 +1,9 @@
 from os import makedirs, environ
-
 from torchinfo import summary
-from torchvision.models import Swin_V2_B_Weights, swin_v2_b
+from torch.nn import Linear,  Sequential, Dropout,  CrossEntropyLoss, Identity,  ReLU
-from torch.nn import Linear, Dropout3d, Sequential, Dropout, Conv2d, CrossEntropyLoss, Identity, MaxPool2d, ReLU, \
-    Softmax
 from torchvision.transforms import Compose, RandomResizedCrop, RandomRotation, ToTensor, \
     RandomHorizontalFlip, \
-    Resize, CenterCrop, RandomAffine, GaussianBlur, RandomAutocontrast, InterpolationMode, AugMix, RandomErasing, \
+    Resize,  RandomAutocontrast, InterpolationMode,  RandomErasing, \
     RandomEqualize, RandomPosterize, RandomPerspective, RandomGrayscale
 import matplotlib
 
@@ -24,11 +21,9 @@ from torch.cuda import is_available
 from torch import no_grad, save, Tensor, load, device
 from datetime import datetime
 from distutils.util import strtobool
-from intel_extension_for_pytorch import optimize
 
 CI = bool(strtobool(environ['CI']))
-# device = device('cuda' if is_available() else 'cpu')
+device = device('cuda' if is_available() else 'cpu')
-device = 'xpu'
 
 model_path: str = join(datadir(), 'artifact', 'vggface2_facenet.pth')
 input_shape: int = 256
@@ -130,7 +125,7 @@ optimizer = Adam(params=[
     {'params': model_gpu[1].parameters(), 'lr': 1e-3},
 ])
 
-model, optimizer = optimize(model=model, optimizer=optimizer)
+# model, optimizer = optimize(model=model, optimizer=optimizer)
 scheduler = lr_scheduler.StepLR(optimizer=optimizer, step_size=10, gamma=0.9)
 epochs = 100
 

finetune/masked_or_not.py

@@ -7,7 +7,7 @@ from torchvision.transforms import Compose, RandomResizedCrop, RandomRotation, T
     RandomHorizontalFlip, \
     Resize, RandomAffine, RandomAdjustSharpness, RandomAutocontrast, RandomEqualize, GaussianBlur, Normalize
 from numpy import arange, ceil, full, float32, uint8, amax, amin
-from torchsummary import summary
+from torchinfo import summary
 from torch.nn import CrossEntropyLoss
 from torch.cuda.amp.grad_scaler import GradScaler
 from torch.cuda.amp import autocast

mask_correct.py

@@ -2,7 +2,6 @@ from torch import zeros, load, no_grad, stack, float32, nn
 from torch.utils.data import DataLoader
 from torchvision.datasets import ImageFolder
 from torchvision.transforms import Compose, functional, ToTensor, Resize, ConvertImageDtype
-from torch2trt import torch2trt
 from PIL import Image
 from numpy import array
 from os import listdir, makedirs

movie_processing/opencv_test.py

@@ -1,8 +1,9 @@
 from cv2 import VideoCapture, getBuildInformation
 from torchvision.models.mobilenetv3 import MobileNetV3
+from settings import datadir
 
 print(getBuildInformation())
-sample_video = VideoCapture('/home/tomokazu/PycharmProjects/helloproject-ai/koi_ing.webm')
+sample_video = VideoCapture('/movie_processing/koi_ing.webm')
 assert sample_video.isOpened()
 
 ret = True

setenv.sh

@@ -0,0 +1,15 @@
+#!/usr/bin/env bash
+SOURCE="${BASH_SOURCE[0]}"
+while [ -h "$SOURCE" ]; do # resolve $SOURCE until the file is no longer a symlink
+  DIR="$( cd -P "$( dirname "$SOURCE" )" && pwd )"
+  SOURCE="$(readlink "$SOURCE")"
+  [[ $SOURCE != /* ]] && SOURCE="$DIR/$SOURCE" # if $SOURCE was a relative symlink, we need to resolve it relative to the path where the symlink file was located
+done
+DIR="$( cd -P "$( dirname "$SOURCE" )" && pwd )"
+
+PYTHONPATH=$PYTHONPATH:$DIR
+
+export PYTHONPATH
+export DATA_DIR="${PYTHONPATH%/}/data"
+
+type conda >/dev/null 2>&1 && conda activate helloproject-ai

settings.py

@@ -1,5 +1,5 @@
 from functools import cache
-from os import getcwd, pardir
+from os import getcwd, pardir, environ
 from os.path import join, abspath, dirname
 
 blog_list = ['angerme-ss-shin', 'angerme-amerika', 'angerme-new', 'juicejuice-official', 'tsubaki-factory',
@@ -53,7 +53,7 @@ def theme_curator(theme: str, blog_id: str) -> str:
 
 @cache
 def datadir():
-    return join('/home/tomokazu/PycharmProjects/helloproject-ai/', 'data')
+    return environ['DATA_DIR']
 
 
 request_header = {

similar_face.py

@@ -1,3 +1,4 @@
+import time
 from shutil import copyfile
 from insightface.app import FaceAnalysis
 from os import getcwd, listdir, makedirs
@@ -15,7 +16,15 @@ if not isdir(argv[2]):
 if not isfile(argv[1]):
     exit(1)
 
-face_analysis = FaceAnalysis()
+face_analysis = FaceAnalysis(providers=[
+    # 'CUDAExecutionProvider',
+    # 'CPUExecutionProvider',
+    ('TensorrtExecutionProvider', {
+        'trt_engine_cache_enable': True,
+        'trt_engine_cache_path': join(getcwd(), 'onnx_cache'),
+        'trt_fp16_enable': True,
+    })
+], allowed_modules=['recognition', 'detection'])
 face_analysis.prepare(ctx_id=0, det_size=(160, 160))
 
 print(argv)
@@ -35,6 +44,7 @@ makedirs(join(getcwd(), dir_name, "true"), exist_ok=True)
 makedirs(join(getcwd(), dir_name, "false"), exist_ok=True)
 
 images = []
+begin = time.time()
 for file in image_files:
     if isfile(join(getcwd(), argv[2], file)):
         # print(join(getcwd(), argv[2], file))
@@ -50,3 +60,5 @@ for file in image_files:
         else:
 
             copyfile(join(getcwd(), argv[2], file), join(getcwd(), dir_name, "false", file))
+
+print(f"{time.time() - begin}sec")

split_train_val.py

@@ -9,7 +9,7 @@ from asyncio import to_thread, gather, run
 from aiofiles import open as a_open
 
 valid_rate = 0.1
-SRC_DIR = join(r'/home/tomokazu/PycharmProjects/helloproject-ai/data/sample_set/')
+SRC_DIR = join(datadir(), 'sample_set')
 DEST_DIR = join(datadir(), 'dataset')
 
 makedirs(DEST_DIR, exist_ok=True)
@@ -51,4 +51,3 @@ with tqdm(listdir(SRC_DIR)) as pbar:
             coroutines.append(co)
         # print(name, file)
         run(waiting(coroutines))
-

test_script/onnx_cacher.py

@@ -0,0 +1,14 @@
+from os import getcwd
+from os.path import join
+from onnxruntime import InferenceSession, SessionOptions
+
+onnx_session = InferenceSession(
+    path_or_bytes="/home/tomokazu/.insightface/models/buffalo_l/w600k_r50.onnx",
+    providers=[
+        ('TensorrtExecutionProvider', {
+            'trt_engine_cache_enable': True,
+            'trt_engine_cache_path': join(getcwd(), 'onnx_cache'),
+            'trt_fp16_enable': True,
+        })
+    ]
+)

test_script/trt_test.py

@@ -0,0 +1,17 @@
+from torch import load, randn, float, half, jit
+import torch_tensorrt
+from torch.nn import Module
+
+model: Module = load(
+    f='/home/tomokazu/PycharmProjects/helloproject-ai/data/artifact/facenet-tl_2023-05-28 23:05:09.874085/model.pth')
+model.cuda()
+model.eval()
+
+example_input = randn(size=[1, 3, 224, 224]).float().cuda()
+
+traced_script_module = jit.trace(model, example_inputs=[example_input])
+tensorrt_module = torch_tensorrt.compile(module=traced_script_module, inputs=[example_input],
+                                         enabled_precisions={float, half},
+                                         truncate_long_and_double=True)
+
+jit.save(tensorrt_module, "trt_test.ts")

torch_dataloader.py

@@ -1,23 +0,0 @@
-from torchvision.transforms import Compose
-import matplotlib.pyplot as plt
-from numpy import transpose
-import torchvision.datasets as datasets
-from torchvision import models
-from PIL import Image
-
-print(dir(datasets))
-
-
-def show_image(x):
-    fig = plt.figure(figsize=(10, 10))
-    for s in range(len(x)):
-        img = x[s].numpy()
-        img = transpose(img, (1, 2, 0))
-        ax1 = fig.add_subplot(1, len(x), s + 1)
-        plt.axis('off')
-        plt.imshow(img)
-
-
-model = models.inception_v3(models.Inception_V3_Weights)
-
-print(model._get_name())

yolo_test.py

@@ -1,5 +0,0 @@
-from super_gradients.training.models import get
-
-yolo_nas = get(model_name='yolo_nas_l', pretrained_weights='coco').cuda()
-
-yolo_nas.predict('橋迫鈴=angerme-new=12687767841-1.jpg', conf=0.8).show()