Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -21,12 +21,14 @@ CHECKPOINTS = {
|
|
| 21 |
|
| 22 |
# 全局变量
|
| 23 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 24 |
-
current_vis = []
|
| 25 |
-
current_bpe = []
|
| 26 |
-
current_index = 0
|
|
|
|
| 27 |
|
| 28 |
def load_model(check_type):
|
| 29 |
-
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
| 30 |
if check_type == 'R50':
|
| 31 |
tokenizer = load_tokenizer('tokenizer_path')
|
| 32 |
model = build_model(argparse.Namespace()).eval()
|
|
@@ -78,6 +80,10 @@ def process_image(model, tokenizer, transform, device, check_type, image, text):
|
|
| 78 |
text_embeds = model.tok_embeddings(input_ids)
|
| 79 |
|
| 80 |
vit_embeds, size1 = model.forward_tokenocr(pixel_values.to(torch.bfloat16).to(device))
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
vit_embeds, size2 = post_process(vit_embeds, target_ratio, check_type)
|
| 82 |
|
| 83 |
# 计算相似度
|
|
@@ -86,26 +92,45 @@ def process_image(model, tokenizer, transform, device, check_type, image, text):
|
|
| 86 |
similarity = text_embeds @ vit_embeds.T
|
| 87 |
resized_size = size1 if size1 is not None else size2
|
| 88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
# 生成可视化
|
| 90 |
attn_map = similarity.reshape(len(text_embeds), resized_size[0], resized_size[1])
|
|
|
|
| 91 |
all_bpe_strings = [tokenizer.decode(input_id) for input_id in input_ids]
|
| 92 |
current_vis = generate_similiarity_map([image], attn_map,
|
| 93 |
[tokenizer.decode([i]) for i in input_ids],
|
| 94 |
[], target_ratio, src_size)
|
| 95 |
|
| 96 |
current_bpe = [tokenizer.decode([i]) for i in input_ids]
|
|
|
|
| 97 |
current_bpe[-1] = text
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
def format_bpe_display(bpe):
|
| 102 |
-
return f"<div style='text-align:center; font-size:20px;'><strong>Current BPE: <span style='color:red;'>{bpe}</span></strong></div>"
|
| 103 |
|
|
|
|
| 104 |
def update_index(change):
|
| 105 |
global current_vis, current_bpe, current_index
|
| 106 |
current_index = max(0, min(len(current_vis) - 1, current_index + change))
|
| 107 |
return current_vis[current_index], format_bpe_display(current_bpe[current_index])
|
| 108 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
# Gradio界面
|
| 110 |
with gr.Blocks(title="BPE Visualization Demo") as demo:
|
| 111 |
gr.Markdown("## BPE Visualization Demo - TokenFD基座模型能力可视化")
|
|
@@ -115,7 +140,7 @@ with gr.Blocks(title="BPE Visualization Demo") as demo:
|
|
| 115 |
model_type = gr.Dropdown(
|
| 116 |
choices=["TokenFD_4096_English_seg", "TokenFD_2048_Bilingual_seg", "R50", "R50_siglip"],
|
| 117 |
label="Select model type",
|
| 118 |
-
value="TokenOCR_4096_English_seg"
|
| 119 |
)
|
| 120 |
image_input = gr.Image(label="Upload images", type="pil")
|
| 121 |
text_input = gr.Textbox(label="Input text")
|
|
@@ -139,34 +164,57 @@ with gr.Blocks(title="BPE Visualization Demo") as demo:
|
|
| 139 |
orig_img = gr.Image(label="Original picture", interactive=False)
|
| 140 |
heatmap = gr.Image(label="BPE visualization", interactive=False)
|
| 141 |
|
| 142 |
-
with gr.Row():
|
| 143 |
-
prev_btn = gr.Button("⬅
|
| 144 |
-
|
|
|
|
| 145 |
|
| 146 |
-
bpe_display = gr.Markdown("Current BPE: ", visible=
|
| 147 |
|
| 148 |
# 事件处理
|
| 149 |
@spaces.GPU
|
| 150 |
def on_run_clicked(model_type, image, text):
|
| 151 |
global current_vis, current_bpe, current_index
|
|
|
|
| 152 |
image, vis, bpe = process_image(*load_model(model_type), model_type, image, text)
|
| 153 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 154 |
|
| 155 |
run_btn.click(
|
| 156 |
on_run_clicked,
|
| 157 |
inputs=[model_type, image_input, text_input],
|
| 158 |
-
outputs=[orig_img, heatmap, bpe_display],
|
|
|
|
|
|
|
|
|
|
|
|
|
| 159 |
)
|
| 160 |
-
|
| 161 |
prev_btn.click(
|
| 162 |
-
lambda: update_index(-1),
|
| 163 |
-
outputs=[heatmap, bpe_display]
|
| 164 |
)
|
| 165 |
-
|
| 166 |
next_btn.click(
|
| 167 |
-
lambda: update_index(1),
|
| 168 |
-
outputs=[heatmap, bpe_display]
|
| 169 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 170 |
|
| 171 |
if __name__ == "__main__":
|
| 172 |
demo.launch()
|
|
|
|
| 21 |
|
| 22 |
# 全局变量
|
| 23 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 24 |
+
current_vis = []
|
| 25 |
+
current_bpe = []
|
| 26 |
+
current_index = 0
|
| 27 |
+
|
| 28 |
|
| 29 |
def load_model(check_type):
|
| 30 |
+
# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 31 |
+
device = torch.device("cuda")
|
| 32 |
if check_type == 'R50':
|
| 33 |
tokenizer = load_tokenizer('tokenizer_path')
|
| 34 |
model = build_model(argparse.Namespace()).eval()
|
|
|
|
| 80 |
text_embeds = model.tok_embeddings(input_ids)
|
| 81 |
|
| 82 |
vit_embeds, size1 = model.forward_tokenocr(pixel_values.to(torch.bfloat16).to(device))
|
| 83 |
+
print("vit_embeds",vit_embeds)
|
| 84 |
+
print("vit_embeds,shape",vit_embeds.shape)
|
| 85 |
+
print("target_ratio",target_ratio)
|
| 86 |
+
print("check_type",check_type)
|
| 87 |
vit_embeds, size2 = post_process(vit_embeds, target_ratio, check_type)
|
| 88 |
|
| 89 |
# 计算相似度
|
|
|
|
| 92 |
similarity = text_embeds @ vit_embeds.T
|
| 93 |
resized_size = size1 if size1 is not None else size2
|
| 94 |
|
| 95 |
+
# print(f"text_embeds shape: {text_embeds.shape}, numel: {text_embeds.numel()}") # text_embeds shape: torch.Size([4, 2048]), numel: 8192
|
| 96 |
+
# print(f"vit_embeds shape: {vit_embeds.shape}, numel: {vit_embeds.numel()}") # vit_embeds shape: torch.Size([9728, 2048]), numel: 19922944
|
| 97 |
+
# print(f"similarity shape: {similarity.shape}, numel: {similarity.numel()}")# similarity shape: torch.Size([4, 9728]), numel: 38912
|
| 98 |
+
|
| 99 |
+
|
| 100 |
# 生成可视化
|
| 101 |
attn_map = similarity.reshape(len(text_embeds), resized_size[0], resized_size[1])
|
| 102 |
+
# attn_map = similarity.reshape(len(text_embeds), *target_ratio)
|
| 103 |
all_bpe_strings = [tokenizer.decode(input_id) for input_id in input_ids]
|
| 104 |
current_vis = generate_similiarity_map([image], attn_map,
|
| 105 |
[tokenizer.decode([i]) for i in input_ids],
|
| 106 |
[], target_ratio, src_size)
|
| 107 |
|
| 108 |
current_bpe = [tokenizer.decode([i]) for i in input_ids]
|
| 109 |
+
# current_bpe[-1] = 'Input text'
|
| 110 |
current_bpe[-1] = text
|
| 111 |
+
print("current_vis",len(current_vis))
|
| 112 |
+
print("current_bpe",len(current_bpe))
|
| 113 |
+
return image, current_vis[0], current_bpe[0]
|
|
|
|
|
|
|
| 114 |
|
| 115 |
+
# 事件处理函数
|
| 116 |
def update_index(change):
|
| 117 |
global current_vis, current_bpe, current_index
|
| 118 |
current_index = max(0, min(len(current_vis) - 1, current_index + change))
|
| 119 |
return current_vis[current_index], format_bpe_display(current_bpe[current_index])
|
| 120 |
|
| 121 |
+
def format_bpe_display(bpe):
|
| 122 |
+
# 使用HTML标签来设置字体大小、颜色,加粗,并居中
|
| 123 |
+
return f"<div style='text-align:center; font-size:20px;'><strong>Current BPE: <span style='color:red;'>{bpe}</span></strong></div>"
|
| 124 |
+
|
| 125 |
+
def update_slider_index(x):
|
| 126 |
+
current_vis = x[1]
|
| 127 |
+
current_bpe = x[2]
|
| 128 |
+
print(f"x: {x[0]}, current_vis length: {len(current_vis)}, current_bpe length: {len(current_bpe)}")
|
| 129 |
+
if 0 <= x < len(current_vis) and 0 <= x < len(current_bpe):
|
| 130 |
+
return current_vis[x], format_bpe_display(current_bpe[x])
|
| 131 |
+
else:
|
| 132 |
+
return None, "索引超出范围"
|
| 133 |
+
|
| 134 |
# Gradio界面
|
| 135 |
with gr.Blocks(title="BPE Visualization Demo") as demo:
|
| 136 |
gr.Markdown("## BPE Visualization Demo - TokenFD基座模型能力可视化")
|
|
|
|
| 140 |
model_type = gr.Dropdown(
|
| 141 |
choices=["TokenFD_4096_English_seg", "TokenFD_2048_Bilingual_seg", "R50", "R50_siglip"],
|
| 142 |
label="Select model type",
|
| 143 |
+
value="TokenOCR_4096_English_seg" # 设置默认值为第一个选项
|
| 144 |
)
|
| 145 |
image_input = gr.Image(label="Upload images", type="pil")
|
| 146 |
text_input = gr.Textbox(label="Input text")
|
|
|
|
| 164 |
orig_img = gr.Image(label="Original picture", interactive=False)
|
| 165 |
heatmap = gr.Image(label="BPE visualization", interactive=False)
|
| 166 |
|
| 167 |
+
with gr.Row() as controls:
|
| 168 |
+
prev_btn = gr.Button("⬅ Last", visible=False)
|
| 169 |
+
index_slider = gr.Slider(0, 1, value=0, step=1, label="BPE index", visible=False)
|
| 170 |
+
next_btn = gr.Button("⮕ Next", visible=False)
|
| 171 |
|
| 172 |
+
bpe_display = gr.Markdown("Current BPE: ", visible=False)
|
| 173 |
|
| 174 |
# 事件处理
|
| 175 |
@spaces.GPU
|
| 176 |
def on_run_clicked(model_type, image, text):
|
| 177 |
global current_vis, current_bpe, current_index
|
| 178 |
+
current_index = 0 # Reset index when new image is processed
|
| 179 |
image, vis, bpe = process_image(*load_model(model_type), model_type, image, text)
|
| 180 |
+
# Update the slider range and set value to 0
|
| 181 |
+
slider_max_val = len(current_bpe) - 1
|
| 182 |
+
bpe_text = format_bpe_display(bpe)
|
| 183 |
+
print("current_vis",len(current_vis))
|
| 184 |
+
print("current_bpe",len(current_bpe))
|
| 185 |
+
return image, vis, bpe_text, slider_max_val
|
| 186 |
|
| 187 |
run_btn.click(
|
| 188 |
on_run_clicked,
|
| 189 |
inputs=[model_type, image_input, text_input],
|
| 190 |
+
outputs=[orig_img, heatmap, bpe_display, index_slider],
|
| 191 |
+
).then(
|
| 192 |
+
lambda max_val: (gr.update(visible=True), gr.update(visible=True, maximum=max_val, value=0), gr.update(visible=True), gr.update(visible=True)),
|
| 193 |
+
inputs=index_slider,
|
| 194 |
+
outputs=[prev_btn, index_slider, next_btn, bpe_display],
|
| 195 |
)
|
| 196 |
+
|
| 197 |
prev_btn.click(
|
| 198 |
+
lambda: (*update_index(-1), current_index),
|
| 199 |
+
outputs=[heatmap, bpe_display, index_slider]
|
| 200 |
)
|
| 201 |
+
|
| 202 |
next_btn.click(
|
| 203 |
+
lambda: (*update_index(1), current_index),
|
| 204 |
+
outputs=[heatmap, bpe_display, index_slider]
|
| 205 |
)
|
| 206 |
+
|
| 207 |
+
# index_slider.change(
|
| 208 |
+
# lambda x: (current_vis[x], format_bpe_display(current_bpe[x])) if 0<=x<len(current_vis else (None,"Invaild")
|
| 209 |
+
# inputs=index_slider,
|
| 210 |
+
# outputs=[heatmap, bpe_display]
|
| 211 |
+
# )
|
| 212 |
+
|
| 213 |
+
index_slider.change(
|
| 214 |
+
update_slider_index,
|
| 215 |
+
inputs=[index_slider,current_vis,current_bpe],
|
| 216 |
+
outputs=[heatmap, bpe_display]
|
| 217 |
+
)
|
| 218 |
|
| 219 |
if __name__ == "__main__":
|
| 220 |
demo.launch()
|