flix-object-detection-wrapper 0.0.1__tar.gz

flix_object_detection_wrapper-0.0.1/PKG-INFO

@@ -0,0 +1,14 @@
+Metadata-Version: 2.4
+Name: flix_object_detection_wrapper
+Version: 0.0.1
+Summary: object_detection_wrapper
+Author: sanchi
+Author-email: sanchi@flixstock.com
+Classifier: Programming Language :: Python :: 3.11
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description
+Dynamic: summary
+
+object_detection_wrapper

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/config.py

@@ -0,0 +1,31 @@
+import os
+
+
+class objDetConfig:
+	def __init__(self, checkpointPath):
+		# checkpoints directory must be supplied by the caller (e.g. the unit test)
+		self.checkpointsPath = checkpointPath
+
+	def getGarmentObjDetectConfig(self):
+		return {
+			"obj_model": os.path.join(self.checkpointsPath, "outfitClipper", "AllGarmentOutfit_step_100000_v4.pth"),
+			"category_count": 2,
+		}
+
+	def getModelObjDetectConfig(self):
+		return {
+			"obj_model": os.path.join(self.checkpointsPath, "modelClipper", "GPDHuman_v4_step_100000.pth"),
+			"category_count": 2,
+		}
+
+	def getShoesObjDetectConfig(self):
+		return {
+			"obj_model": os.path.join(self.checkpointsPath, "shoesClipper", "Model_shoe_step_032500_v4.pth"),
+			"category_count": 2,
+		}
+
+	def getHairObjDetectConfig(self):
+		return {
+			"obj_model": os.path.join(self.checkpointsPath, "hairClipper", "hair_v4_step_057500.pth"),
+			"category_count": 2,
+		}

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/main.py

@@ -0,0 +1,110 @@
+import os
+
+import cv2
+import numpy as np
+from termcolor import colored
+
+# try:
+# 	from objdetect_pytorch import Bolt
+# except ImportError:
+# 	from .objdetect_pytorch import Bolt
+
+from flix_clip2.objdetect_pytorch import Bolt
+
+def initialize_detector(checkpoint_dict, gpu="0"):
+	"""Build the object-detector object once, so it can be reused across many images.
+
+	checkpoint_dict: {"obj_model": <path.pth>, "category_count": <int>} (from any *ObjDetectConfig).
+	gpu: gpu id as a string; "-1" for CPU.
+	Returns the initialized detector (Bolt) object.
+	"""
+	assert os.path.isfile(checkpoint_dict["obj_model"]), \
+		"Checkpoint not found: %s" % checkpoint_dict["obj_model"]
+	detector = Bolt(checkpoint_dict, gpu)
+	return detector
+
+
+def pad_bbox(box, img_w, img_h, height_percent=0.20, width_percent=0.20):
+	"""Expand a detection box by a fraction of its own size, clamped to image bounds.
+
+	height_percent: fraction of the bbox height added to BOTH top and bottom.
+	width_percent:  fraction of the bbox width  added to BOTH left and right.
+	Returns (xmin, ymin, xmax, ymax).
+	"""
+	bbox_w = box['xmax'] - box['xmin']
+	bbox_h = box['ymax'] - box['ymin']
+	pad_x = int(width_percent * bbox_w)
+	pad_y = int(height_percent * bbox_h)
+	xmin = max(0, box['xmin'] - pad_x)
+	ymin = max(0, box['ymin'] - pad_y)
+	xmax = min(img_w, box['xmax'] + pad_x)
+	ymax = min(img_h, box['ymax'] + pad_y)
+	return xmin, ymin, xmax, ymax
+
+
+def run_inference(detector, input_dict):
+	"""Run detection on one image, save all outputs, and return the first (top) bbox.
+
+	This is detector-agnostic — identical whether the object came from the garment,
+	model, shoes, ... config. Only the input_dict carries per-call parameters, so new
+	parameters can be added later without changing the signature.
+
+	input_dict keys:
+		image_path     (required) path to the image to detect on
+		output_dir     (required) directory to write the crops/masks into
+		height_percent (optional, default 0.20) padding fraction, top+bottom
+		width_percent  (optional, default 0.20) padding fraction, left+right
+
+	For each detection i it saves:
+		<i>.jpg       crop exactly as returned by detector.run
+		<i>_pad.jpg   crop expanded by (height_percent, width_percent)
+		<i>_mask.jpg  full-size grayscale mask: padded region = 255, background = 0
+
+	Returns the first bbox dict {'xmin','ymin','xmax','ymax','score','label'}, or None
+	if nothing was detected.
+	"""
+	image_path = input_dict['image_path']
+	output_dir = input_dict['output_dir']
+	height_percent = input_dict.get('height_percent', 0.20)
+	width_percent = input_dict.get('width_percent', 0.20)
+
+	assert os.path.isfile(image_path), "Image not found: %s" % image_path
+	os.makedirs(output_dir, exist_ok=True)
+
+	# ------------ inference ------------
+	flag, crop_img_list, detection_box_list = detector.run({"filepath": image_path}, MARGIN=0)
+	if not flag or len(detection_box_list) == 0:
+		print(colored("Detection FAILED: %s" % image_path, "red"))
+		return None
+
+	# original image size — masks are full-size grayscale of the detection input
+	orig_img = cv2.imread(image_path, 1)
+	orig_h, orig_w = orig_img.shape[:2]
+
+	# ------------ save all outputs ------------
+	for i, box in enumerate(detection_box_list):
+		print(box['xmin'], box['ymin'], box['xmax'], box['ymax'], box['label'], box['score'])
+
+		# 1. crop exactly as returned by detector.run
+		crop_path = os.path.join(output_dir, "%d.jpg" % i)
+		cv2.imwrite(crop_path, crop_img_list[i])
+		print(colored("  crop saved: %s" % crop_path, "cyan"))
+
+		# padded box (clamped to image bounds)
+		px_min, py_min, px_max, py_max = pad_bbox(box, orig_w, orig_h, height_percent, width_percent)
+
+		# 2. padded crop
+		pad_crop = orig_img[py_min:py_max, px_min:px_max]
+		pad_crop_path = os.path.join(output_dir, "%d_pad.jpg" % i)
+		cv2.imwrite(pad_crop_path, pad_crop)
+		print(colored("  padded crop saved: %s" % pad_crop_path, "cyan"))
+
+		# 3. full-size grayscale mask of the PADDED region: white (255) inside, black (0) outside
+		mask = np.zeros((orig_h, orig_w), dtype=np.uint8)
+		mask[py_min:py_max, px_min:px_max] = 255
+		mask_path = os.path.join(output_dir, "%d_mask.jpg" % i)
+		cv2.imwrite(mask_path, mask)
+		print(colored("  mask saved: %s" % mask_path, "cyan"))
+
+	# returns the first (top-confidence) bbox
+	return detection_box_list[0]

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/setup.py

@@ -0,0 +1,16 @@
+import setuptools
+
+setuptools.setup(
+    name="flix_object_detection_wrapper",
+    version="0.0.1",
+    author="sanchi",
+    author_email="sanchi@flixstock.com",
+    description="object_detection_wrapper",
+    long_description='object_detection_wrapper',
+    packages=setuptools.find_packages(),
+    include_package_data=True,
+    classifiers=[
+        "Programming Language :: Python :: 3.11",
+        # "Operating System :: Ubuntu 20.04",
+    ],
+)

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/unit_test.py

@@ -0,0 +1,42 @@
+import os
+
+from termcolor import colored
+from flix_object_detection_wrapper.main import initialize_detector, run_inference
+from flix_object_detection_wrapper.config import objDetConfig
+
+# try:
+# 	from main import initialize_detector, run_inference
+# 	from config import objDetConfig
+# except ImportError:
+# 	from .main import initialize_detector, run_inference
+# 	from .config import objDetConfig
+
+if __name__ == "__main__":
+	gpu = "0"  # use "-1" for CPU
+
+	# module-local checkpoints, so object_det/ works wherever it is copied
+	checkpoints_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/checkpoints"
+	img_path = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/input/aleksandra_18998.jpg"
+	out_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/output/aleksandra_18998_2"
+
+	# ------------ config ------------
+	# reads {"obj_model": <path.pth>, "category_count": <int>}; swap the getter to use another detector
+	checkpoint_dict = objDetConfig(checkpoints_dir).getGarmentObjDetectConfig()
+	print(colored("Config: %s" % checkpoint_dict, "yellow"))
+	print(colored("Image : %s" % img_path, "yellow"))
+
+	# ------------ initialize once (reuse across many images) ------------
+	detector = initialize_detector(checkpoint_dict, gpu)
+
+	# ------------ infer ------------
+	# per-call parameters go in input_dict, so new ones can be added later without signature changes
+	input_dict = {
+		"image_path": img_path,
+		"output_dir": out_dir,
+		"height_percent": 0.01,
+		"width_percent": 0.01,
+	}
+	first_bbox = run_inference(detector, input_dict)
+
+	assert first_bbox is not None, "Detection FAILED — no boxes returned"
+	print(colored("PASS: first bbox = %s" % first_bbox, "green"))

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/unit_test_hair.py

@@ -0,0 +1,43 @@
+import os
+
+from termcolor import colored
+
+# try:
+# 	from main import initialize_detector, run_inference
+# 	from config import objDetConfig
+# except ImportError:
+# 	from .main import initialize_detector, run_inference
+# 	from .config import objDetConfig
+
+from flix_object_detection_wrapper.main import initialize_detector, run_inference
+from flix_object_detection_wrapper.config import objDetConfig
+
+if __name__ == "__main__":
+	gpu = "0"  # use "-1" for CPU
+
+	# module-local checkpoints, so object_det/ works wherever it is copied
+	checkpoints_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/checkpoints"
+	img_path = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/input/aleksandra_18998.jpg"
+	out_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/output/aleksandra_18998_hair"
+
+	# ------------ config ------------
+	# hair detector; only this line differs from the garment unit test
+	checkpoint_dict = objDetConfig(checkpoints_dir).getHairObjDetectConfig()
+	print(colored("Config: %s" % checkpoint_dict, "yellow"))
+	print(colored("Image : %s" % img_path, "yellow"))
+
+	# ------------ initialize once (reuse across many images) ------------
+	detector = initialize_detector(checkpoint_dict, gpu)
+
+	# ------------ infer ------------
+	# per-call parameters go in input_dict, so new ones can be added later without signature changes
+	input_dict = {
+		"image_path": img_path,
+		"output_dir": out_dir,
+		"height_percent": 0.20,
+		"width_percent": 0.20,
+	}
+	first_bbox = run_inference(detector, input_dict)
+
+	assert first_bbox is not None, "Detection FAILED — no boxes returned"
+	print(colored("PASS: first bbox = %s" % first_bbox, "green"))

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/unit_test_model.py

@@ -0,0 +1,43 @@
+import os
+
+from termcolor import colored
+
+# try:
+# 	from main import initialize_detector, run_inference
+# 	from config import objDetConfig
+# except ImportError:
+# 	from .main import initialize_detector, run_inference
+# 	from .config import objDetConfig
+
+from flix_object_detection_wrapper.main import initialize_detector, run_inference
+from flix_object_detection_wrapper.config import objDetConfig
+
+if __name__ == "__main__":
+	gpu = "0"  # use "-1" for CPU
+
+	# module-local checkpoints, so object_det/ works wherever it is copied
+	checkpoints_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/checkpoints"
+	img_path = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/input/aleksandra_18998.jpg"
+	out_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/output/aleksandra_18998_model"
+
+	# ------------ config ------------
+	# model (human) detector; only this line differs from the garment unit test
+	checkpoint_dict = objDetConfig(checkpoints_dir).getModelObjDetectConfig()
+	print(colored("Config: %s" % checkpoint_dict, "yellow"))
+	print(colored("Image : %s" % img_path, "yellow"))
+
+	# ------------ initialize once (reuse across many images) ------------
+	detector = initialize_detector(checkpoint_dict, gpu)
+
+	# ------------ infer ------------
+	# per-call parameters go in input_dict, so new ones can be added later without signature changes
+	input_dict = {
+		"image_path": img_path,
+		"output_dir": out_dir,
+		"height_percent": 0.01,
+		"width_percent": 0.01,
+	}
+	first_bbox = run_inference(detector, input_dict)
+
+	assert first_bbox is not None, "Detection FAILED — no boxes returned"
+	print(colored("PASS: first bbox = %s" % first_bbox, "green"))

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper/unit_test_shoes.py

@@ -0,0 +1,43 @@
+import os
+
+from termcolor import colored
+
+# try:
+# 	from main import initialize_detector, run_inference
+# 	from config import objDetConfig
+# except ImportError:
+# 	from .main import initialize_detector, run_inference
+# 	from .config import objDetConfig
+
+from flix_object_detection_wrapper.main import initialize_detector, run_inference
+from flix_object_detection_wrapper.config import objDetConfig
+
+if __name__ == "__main__":
+	gpu = "0"  # use "-1" for CPU
+
+	# module-local checkpoints, so object_det/ works wherever it is copied
+	checkpoints_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/checkpoints"
+	img_path = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/input/aleksandra_18998.jpg"
+	out_dir = "/home/ubuntu/projects/python/sanchi/clippers/object_detection_wrapper/data/test_object_detection/output/aleksandra_18998_shoes"
+
+	# ------------ config ------------
+	# shoes detector; only this line differs from the garment unit test
+	checkpoint_dict = objDetConfig(checkpoints_dir).getShoesObjDetectConfig()
+	print(colored("Config: %s" % checkpoint_dict, "yellow"))
+	print(colored("Image : %s" % img_path, "yellow"))
+
+	# ------------ initialize once (reuse across many images) ------------
+	detector = initialize_detector(checkpoint_dict, gpu)
+
+	# ------------ infer ------------
+	# per-call parameters go in input_dict, so new ones can be added later without signature changes
+	input_dict = {
+		"image_path": img_path,
+		"output_dir": out_dir,
+		"height_percent": 0.01,
+		"width_percent": 0.01,
+	}
+	first_bbox = run_inference(detector, input_dict)
+
+	assert first_bbox is not None, "Detection FAILED — no boxes returned"
+	print(colored("PASS: first bbox = %s" % first_bbox, "green"))

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper.egg-info/PKG-INFO

@@ -0,0 +1,14 @@
+Metadata-Version: 2.4
+Name: flix_object_detection_wrapper
+Version: 0.0.1
+Summary: object_detection_wrapper
+Author: sanchi
+Author-email: sanchi@flixstock.com
+Classifier: Programming Language :: Python :: 3.11
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description
+Dynamic: summary
+
+object_detection_wrapper

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper.egg-info/SOURCES.txt

@@ -0,0 +1,13 @@
+setup.py
+flix_object_detection_wrapper/__init__.py
+flix_object_detection_wrapper/config.py
+flix_object_detection_wrapper/main.py
+flix_object_detection_wrapper/setup.py
+flix_object_detection_wrapper/unit_test.py
+flix_object_detection_wrapper/unit_test_hair.py
+flix_object_detection_wrapper/unit_test_model.py
+flix_object_detection_wrapper/unit_test_shoes.py
+flix_object_detection_wrapper.egg-info/PKG-INFO
+flix_object_detection_wrapper.egg-info/SOURCES.txt
+flix_object_detection_wrapper.egg-info/dependency_links.txt
+flix_object_detection_wrapper.egg-info/top_level.txt

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

flix_object_detection_wrapper-0.0.1/flix_object_detection_wrapper.egg-info/top_level.txt

@@ -0,0 +1 @@
+flix_object_detection_wrapper

flix_object_detection_wrapper-0.0.1/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

flix_object_detection_wrapper-0.0.1/setup.py

@@ -0,0 +1,16 @@
+import setuptools
+
+setuptools.setup(
+    name="flix_object_detection_wrapper",
+    version="0.0.1",
+    author="sanchi",
+    author_email="sanchi@flixstock.com",
+    description="object_detection_wrapper",
+    long_description='object_detection_wrapper',
+    packages=setuptools.find_packages(),
+    include_package_data=True,
+    classifiers=[
+        "Programming Language :: Python :: 3.11",
+        # "Operating System :: Ubuntu 20.04",
+    ],
+)