Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,20 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from PIL import Image, ImageDraw
|
| 3 |
import torch
|
| 4 |
-
from transformers import OwlViTProcessor, OwlViTForObjectDetection
|
| 5 |
from transformers.image_transforms import center_to_corners_format
|
| 6 |
from transformers.models.owlvit.modeling_owlvit import box_iou
|
| 7 |
from functools import partial
|
|
|
|
| 8 |
|
|
|
|
| 9 |
|
| 10 |
processor = OwlViTProcessor.from_pretrained("google/owlvit-base-patch32")
|
| 11 |
model = OwlViTForObjectDetection.from_pretrained("google/owlvit-base-patch32")
|
| 12 |
|
| 13 |
-
from transformers.models.owlvit.modeling_owlvit import OwlViTImageGuidedObjectDetectionOutput
|
|
|
|
|
|
|
| 14 |
|
| 15 |
|
| 16 |
|
|
@@ -24,9 +28,7 @@ def classpredictionhead_box_forward(
|
|
| 24 |
|
| 25 |
# Normalize image and text features
|
| 26 |
image_class_embeds = image_class_embeds / (torch.linalg.norm(image_class_embeds, dim=-1, keepdim=True) + 1e-6)
|
| 27 |
-
print(image_class_embeds.shape)
|
| 28 |
query_embeds = image_class_embeds[0, query_indice].unsqueeze(0).unsqueeze(0)
|
| 29 |
-
print(query_embeds.shape)
|
| 30 |
# query_embeds = query_embeds / (torch.linalg.norm(query_embeds, dim=-1, keepdim=True) + 1e-6)
|
| 31 |
|
| 32 |
# Get class predictions
|
|
@@ -66,6 +68,8 @@ def class_predictor(
|
|
| 66 |
|
| 67 |
|
| 68 |
|
|
|
|
|
|
|
| 69 |
def get_max_iou_indice(target_pred_boxes, query_box, target_sizes):
|
| 70 |
boxes = center_to_corners_format(target_pred_boxes)
|
| 71 |
img_h, img_w = target_sizes.unbind(1)
|
|
@@ -104,6 +108,12 @@ def box_guided_detection(
|
|
| 104 |
batch_size, num_patches, num_patches, hidden_dim = feature_map.shape
|
| 105 |
image_feats = torch.reshape(feature_map, (batch_size, num_patches * num_patches, hidden_dim))
|
| 106 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
target_pred_boxes = self.box_predictor(image_feats, feature_map)
|
| 108 |
|
| 109 |
# Get MAX IOU box corresponding embedding
|
|
@@ -113,6 +123,9 @@ def box_guided_detection(
|
|
| 113 |
(pred_logits, class_embeds) = self.class_predictor(image_feats=image_feats, query_indice=query_indice)
|
| 114 |
|
| 115 |
|
|
|
|
|
|
|
|
|
|
| 116 |
if not return_dict:
|
| 117 |
output = (
|
| 118 |
feature_map,
|
|
@@ -150,31 +163,30 @@ def prepare_embedds(xmin, ymin, xmax, ymax, image):
|
|
| 150 |
|
| 151 |
def manul_box_change(xmin, ymin, xmax, ymax, image):
|
| 152 |
box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 153 |
-
return (image, [(box, "manul")])
|
| 154 |
|
| 155 |
def threshold_change(xmin, ymin, xmax, ymax, image, threshold, nms):
|
| 156 |
manul_box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 157 |
|
| 158 |
global outputs
|
| 159 |
-
target_sizes = torch.Tensor([image.size[::-1]])
|
| 160 |
|
| 161 |
results = processor.post_process_image_guided_detection(outputs=outputs, threshold=threshold, nms_threshold=nms, target_sizes=target_sizes)
|
| 162 |
|
| 163 |
boxes = results[0]['boxes'].type(torch.int64).tolist()
|
| 164 |
scores = results[0]['scores'].tolist()
|
| 165 |
labels = list(zip(boxes, scores))
|
| 166 |
-
labels.append((manul_box, "manual"))
|
| 167 |
|
| 168 |
cnt = len(boxes)
|
| 169 |
|
| 170 |
-
return (image, labels), cnt
|
| 171 |
|
| 172 |
def one_shot_detect(xmin, ymin, xmax, ymax, image, threshold, nms):
|
| 173 |
manul_box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 174 |
|
| 175 |
global outputs
|
| 176 |
-
target_sizes = torch.Tensor([image.size[::-1]])
|
| 177 |
-
inputs = processor(images=image.convert("RGB"), return_tensors="pt")
|
| 178 |
outputs = model.box_guided_detection(**inputs, query_box=torch.Tensor([manul_box]), target_sizes=target_sizes)
|
| 179 |
|
| 180 |
results = processor.post_process_image_guided_detection(outputs=outputs, threshold=threshold, nms_threshold=nms, target_sizes=target_sizes)
|
|
@@ -182,37 +194,60 @@ def one_shot_detect(xmin, ymin, xmax, ymax, image, threshold, nms):
|
|
| 182 |
boxes = results[0]['boxes'].type(torch.int64).tolist()
|
| 183 |
scores = results[0]['scores'].tolist()
|
| 184 |
labels = list(zip(boxes, scores))
|
| 185 |
-
labels.append((manul_box, "manual"))
|
| 186 |
|
| 187 |
cnt = len(boxes)
|
| 188 |
|
| 189 |
-
return (image, labels), cnt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 190 |
|
| 191 |
|
| 192 |
with gr.Blocks() as demo:
|
| 193 |
with gr.Row():
|
| 194 |
with gr.Column():
|
| 195 |
-
|
|
|
|
| 196 |
threshold = gr.Number(0.95, label="threshold", step=0.01)
|
| 197 |
nms = gr.Number(0.3, label="nms", step=0.01)
|
| 198 |
cnt = gr.Number(0, label="count", interactive=False)
|
| 199 |
with gr.Column():
|
| 200 |
annotatedimage = gr.AnnotatedImage()
|
| 201 |
with gr.Row():
|
| 202 |
-
xmin = gr.Number(
|
| 203 |
-
ymin = gr.Number(
|
| 204 |
-
xmax = gr.Number(
|
| 205 |
-
ymax = gr.Number(
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 216 |
|
| 217 |
|
| 218 |
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from PIL import Image, ImageDraw
|
| 3 |
import torch
|
| 4 |
+
from transformers import OwlViTProcessor, OwlViTForObjectDetection, OwlViTModel, OwlViTImageProcessor
|
| 5 |
from transformers.image_transforms import center_to_corners_format
|
| 6 |
from transformers.models.owlvit.modeling_owlvit import box_iou
|
| 7 |
from functools import partial
|
| 8 |
+
import numpy as np
|
| 9 |
|
| 10 |
+
# from utils import iou
|
| 11 |
|
| 12 |
processor = OwlViTProcessor.from_pretrained("google/owlvit-base-patch32")
|
| 13 |
model = OwlViTForObjectDetection.from_pretrained("google/owlvit-base-patch32")
|
| 14 |
|
| 15 |
+
from transformers.models.owlvit.modeling_owlvit import OwlViTImageGuidedObjectDetectionOutput, OwlViTClassPredictionHead
|
| 16 |
+
|
| 17 |
+
|
| 18 |
|
| 19 |
|
| 20 |
|
|
|
|
| 28 |
|
| 29 |
# Normalize image and text features
|
| 30 |
image_class_embeds = image_class_embeds / (torch.linalg.norm(image_class_embeds, dim=-1, keepdim=True) + 1e-6)
|
|
|
|
| 31 |
query_embeds = image_class_embeds[0, query_indice].unsqueeze(0).unsqueeze(0)
|
|
|
|
| 32 |
# query_embeds = query_embeds / (torch.linalg.norm(query_embeds, dim=-1, keepdim=True) + 1e-6)
|
| 33 |
|
| 34 |
# Get class predictions
|
|
|
|
| 68 |
|
| 69 |
|
| 70 |
|
| 71 |
+
|
| 72 |
+
|
| 73 |
def get_max_iou_indice(target_pred_boxes, query_box, target_sizes):
|
| 74 |
boxes = center_to_corners_format(target_pred_boxes)
|
| 75 |
img_h, img_w = target_sizes.unbind(1)
|
|
|
|
| 108 |
batch_size, num_patches, num_patches, hidden_dim = feature_map.shape
|
| 109 |
image_feats = torch.reshape(feature_map, (batch_size, num_patches * num_patches, hidden_dim))
|
| 110 |
|
| 111 |
+
# batch_size, num_patches, num_patches, hidden_dim = query_feature_map.shape
|
| 112 |
+
# query_image_feats = torch.reshape(query_feature_map, (batch_size, num_patches * num_patches, hidden_dim))
|
| 113 |
+
# # Get top class embedding and best box index for each query image in batch
|
| 114 |
+
# query_embeds, best_box_indices, query_pred_boxes = self.embed_image_query(query_image_feats, query_feature_map)
|
| 115 |
+
|
| 116 |
+
# Predict object boxes
|
| 117 |
target_pred_boxes = self.box_predictor(image_feats, feature_map)
|
| 118 |
|
| 119 |
# Get MAX IOU box corresponding embedding
|
|
|
|
| 123 |
(pred_logits, class_embeds) = self.class_predictor(image_feats=image_feats, query_indice=query_indice)
|
| 124 |
|
| 125 |
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
|
| 129 |
if not return_dict:
|
| 130 |
output = (
|
| 131 |
feature_map,
|
|
|
|
| 163 |
|
| 164 |
def manul_box_change(xmin, ymin, xmax, ymax, image):
|
| 165 |
box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 166 |
+
return (image["image"], [(box, "manul")])
|
| 167 |
|
| 168 |
def threshold_change(xmin, ymin, xmax, ymax, image, threshold, nms):
|
| 169 |
manul_box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 170 |
|
| 171 |
global outputs
|
| 172 |
+
target_sizes = torch.Tensor([image["image"].size[::-1]])
|
| 173 |
|
| 174 |
results = processor.post_process_image_guided_detection(outputs=outputs, threshold=threshold, nms_threshold=nms, target_sizes=target_sizes)
|
| 175 |
|
| 176 |
boxes = results[0]['boxes'].type(torch.int64).tolist()
|
| 177 |
scores = results[0]['scores'].tolist()
|
| 178 |
labels = list(zip(boxes, scores))
|
|
|
|
| 179 |
|
| 180 |
cnt = len(boxes)
|
| 181 |
|
| 182 |
+
return (image["image"], labels), cnt
|
| 183 |
|
| 184 |
def one_shot_detect(xmin, ymin, xmax, ymax, image, threshold, nms):
|
| 185 |
manul_box = (int(xmin), int(ymin), int(xmax), int(ymax))
|
| 186 |
|
| 187 |
global outputs
|
| 188 |
+
target_sizes = torch.Tensor([image["image"].size[::-1]])
|
| 189 |
+
inputs = processor(images=image["image"].convert("RGB"), return_tensors="pt")
|
| 190 |
outputs = model.box_guided_detection(**inputs, query_box=torch.Tensor([manul_box]), target_sizes=target_sizes)
|
| 191 |
|
| 192 |
results = processor.post_process_image_guided_detection(outputs=outputs, threshold=threshold, nms_threshold=nms, target_sizes=target_sizes)
|
|
|
|
| 194 |
boxes = results[0]['boxes'].type(torch.int64).tolist()
|
| 195 |
scores = results[0]['scores'].tolist()
|
| 196 |
labels = list(zip(boxes, scores))
|
|
|
|
| 197 |
|
| 198 |
cnt = len(boxes)
|
| 199 |
|
| 200 |
+
return (image["image"], labels), cnt
|
| 201 |
+
|
| 202 |
+
def save_embedding(exam):
|
| 203 |
+
print(exam)
|
| 204 |
+
global outputs
|
| 205 |
+
embedding = outputs["class_embeds"][0, outputs["logits"].argmax()]
|
| 206 |
+
return embedding.detach().numpy()
|
| 207 |
+
|
| 208 |
+
|
| 209 |
+
def sketch2box(sketch_box):
|
| 210 |
+
mask = sketch_box["mask"].convert("L")
|
| 211 |
+
mask = np.array(mask)
|
| 212 |
+
|
| 213 |
+
masked_index = np.where(mask == 255)
|
| 214 |
+
if len(masked_index[0]) == 0:
|
| 215 |
+
return (sketch_box["image"], []), -1, -1, -1, -1
|
| 216 |
+
xmin, ymin, xmax, ymax = masked_index[1].min(), masked_index[0].min(), masked_index[1].max(), masked_index[0].max()
|
| 217 |
+
box = (xmin, ymin, xmax, ymax)
|
| 218 |
+
|
| 219 |
+
return (sketch_box["image"], [(box, "manual")]), xmin, ymin, xmax, ymax
|
| 220 |
|
| 221 |
|
| 222 |
with gr.Blocks() as demo:
|
| 223 |
with gr.Row():
|
| 224 |
with gr.Column():
|
| 225 |
+
sketch_box = gr.Image(type="pil", source="upload", tool="sketch")
|
| 226 |
+
box_preview = gr.AnnotatedImage(type="pil", interactive=False, height=256)
|
| 227 |
threshold = gr.Number(0.95, label="threshold", step=0.01)
|
| 228 |
nms = gr.Number(0.3, label="nms", step=0.01)
|
| 229 |
cnt = gr.Number(0, label="count", interactive=False)
|
| 230 |
with gr.Column():
|
| 231 |
annotatedimage = gr.AnnotatedImage()
|
| 232 |
with gr.Row():
|
| 233 |
+
xmin = gr.Number(-1, label="xmin")
|
| 234 |
+
ymin = gr.Number(-1, label="ymin")
|
| 235 |
+
xmax = gr.Number(-1, label="xmax")
|
| 236 |
+
ymax = gr.Number(-1, label="ymax")
|
| 237 |
+
with gr.Row():
|
| 238 |
+
run_button = gr.Button(variant="primary")
|
| 239 |
+
# save_button = gr.Button("Save", variant="secondary")
|
| 240 |
+
|
| 241 |
+
|
| 242 |
+
sketch_box.edit(sketch2box, [sketch_box], [box_preview, xmin, ymin, xmax, ymax])
|
| 243 |
+
xmin.change(manul_box_change, [xmin, ymin, xmax, ymax, sketch_box], [box_preview])
|
| 244 |
+
ymin.change(manul_box_change, [xmin, ymin, xmax, ymax, sketch_box], [box_preview])
|
| 245 |
+
xmax.change(manul_box_change, [xmin, ymin, xmax, ymax, sketch_box], [box_preview])
|
| 246 |
+
ymax.change(manul_box_change, [xmin, ymin, xmax, ymax, sketch_box], [box_preview])
|
| 247 |
+
threshold.change(threshold_change, [xmin, ymin, xmax, ymax, sketch_box, threshold, nms], [annotatedimage, cnt])
|
| 248 |
+
nms.change(threshold_change, [xmin, ymin, xmax, ymax, sketch_box, threshold, nms], [annotatedimage, cnt])
|
| 249 |
+
run_button.click(one_shot_detect, [xmin, ymin, xmax, ymax, sketch_box, threshold, nms], [annotatedimage, cnt])
|
| 250 |
+
|
| 251 |
|
| 252 |
|
| 253 |
|