Update app.py

app.py

@@ -12,22 +12,99 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
 
 model.eval()
 
+def validate_input(title, abstract):
+    """验证输入是否符合要求"""
+
+    # 黑名单：屏蔽非拉丁字符
+    non_latin_pattern = re.compile(r'[^\u0000-\u007F]')
+    if len(title.split(' '))<4:
+        return False, "The title must be at least 3 words long."
+    if len(abstract.split(' ')) < 50:
+        return False, "The abstract must be at least 50 words long."
+    if len((title + abstract).split(' '))>1024:
+        return True, "Warning, The input length is approaching tokenization limits (1024) and may be truncated without further warning!"
+    if non_latin_pattern.search(title):
+        return False, "The title contains invalid characters. Only English letters and special symbols are allowed."
+    if non_latin_pattern.search(abstract):
+        return False, "The abstract contains invalid characters. Only English letters and special symbols are allowed."
+
+    return True, "Inputs are valid! Good to go!"
+
+def update_button_status(title, abstract):
+    """根据输入内容动态更新按钮状态"""
+    valid, message = validate_input(title, abstract)
+    if not valid:
+        return gr.update(value="Error: " + message), gr.update(interactive=False)
+    return gr.update(value=message), gr.update(interactive=True)
+
+
 def predict(title, abstract):
-    # 将标题和摘要处理为一个单一的字符串
     text = f'''Given a certain paper, Title: {title}\n Abstract: {abstract}. \n Predict its normalized academic impact (between 0 and 1):'''
     inputs = tokenizer(text, return_tensors="pt")
     with torch.no_grad():
         outputs = model(**inputs.to(device))
-    # 应用 Sigmoid 函数来获取概率输出
     probability = torch.sigmoid(outputs.logits).item()
-    return {"Predicted Impact": round(probability, 4)}
+    # reason for +0.05: We observed that the predicted values in the web demo are generally around 0.05 lower than those in the local deployment (due to differences in software/hardware environments). Therefore, we applied the following compensation in the web demo. Please do not use this in the local deployment.
+    return round(probability + 0.05, 4)
+
+
+# 示例数据
+examples = [
+    [
+        "LoRA-IR: Taming Low-Rank Experts for Efficient All-in-One Image Restoration",
+        ('''Prompt-based all-in-one image restoration (IR) frameworks have achieved
+ remarkable performance by incorporating degradation-specific information into
+ prompt modules. Nevertheless, handling the complex and diverse degradations
+ encountered in real-world scenarios remains a significant challenge. To address
+ this challenge, we propose LoRA-IR, a flexible framework that dynamically
+ leverages compact low-rank experts to facilitate efficient all-in-one image
+ restoration. Specifically, LoRA-IR consists of two training stages:
+ degradation-guided pre-training and parameter-efficient fine-tuning. In the
+ pre-training stage, we enhance the pre-trained CLIP model by introducing a
+ simple mechanism that scales it to higher resolutions, allowing us to extract
+ robust degradation representations that adaptively guide the IR network. In the
+ fine-tuning stage, we refine the pre-trained IR network using low-rank
+ adaptation (LoRA). Built upon a Mixture-of-Experts (MoE) architecture, LoRA-IR
+ dynamically integrates multiple low-rank restoration experts through a
+ degradation-guided router. This dynamic integration mechanism significantly
+ enhances our model's adaptability to diverse and unknown degradations in
+ complex real-world scenarios. Extensive experiments demonstrate that LoRA-IR
+ achieves state-of-the-art performance across 14 image restoration tasks and 29
+ benchmarks. Code and pre-trained models will be available at:
+ https://github.com/shallowdream204/LoRA-IR.''')
+    ],
+    [
+        "ConsistentAvatar: Learning to Diffuse Fully Consistent Talking Head Avatar with Temporal Guidance",
+        ('''Diffusion models have shown impressive potential on talking head generation.
+ While plausible appearance and talking effect are achieved, these methods still
+ suffer from temporal, 3D or expression inconsistency due to the error
+ accumulation and inherent limitation of single-image generation ability. In
+ this paper, we propose ConsistentAvatar, a novel framework for fully consistent
+ and high-fidelity talking avatar generation. Instead of directly employing
+ multi-modal conditions to the diffusion process, our method learns to first
+ model the temporal representation for stability between adjacent frames.
+ Specifically, we propose a Temporally-Sensitive Detail (TSD) map containing
+ high-frequency feature and contours that vary significantly along the time
+ axis. Using a temporal consistent diffusion module, we learn to align TSD of
+ the initial result to that of the video frame ground truth. The final avatar is
+ generated by a fully consistent diffusion module, conditioned on the aligned
+ TSD, rough head normal, and emotion prompt embedding. We find that the aligned
+ TSD, which represents the temporal patterns, constrains the diffusion process
+ to generate temporally stable talking head. Further, its reliable guidance
+ complements the inaccuracy of other conditions, suppressing the accumulated
+ error while improving the consistency on various aspects. Extensive experiments
+ demonstrate that ConsistentAvatar outperforms the state-of-the-art methods on
+ the generated appearance, 3D, expression and temporal consistency. Project
+ page: https://njust-yang.github.io/ConsistentAvatar.github.io/''')
+    ]，
+
+]
 
 # 创建 Gradio 界面
 with gr.Blocks() as iface:
     gr.Markdown("""
-    # 🧠 Predict Academic Impact of newly published paper!!
-    ### Estimate the future academic impact of a paper using AI
-    ---
+    # 🧠 Predict Academic Impact of Newly Published Paper!
+    ### Estimate the future academic impact of a paper using LLM
     [Read the full paper](https://arxiv.org/abs/2408.03934)
     """)
     with gr.Row():
@@ -42,21 +119,45 @@ with gr.Blocks() as iface:
                 placeholder="Enter Paper Abstract Here... (Do not input line breaks. No more than 1024 tokens.)",
                 label="Paper Abstract"
             )
-            submit_button = gr.Button("Predict Impact")
+            validation_status = gr.Textbox(label="Validation Status", interactive=False)
+            submit_button = gr.Button("Predict Impact", interactive=False)
         with gr.Column():
             output = gr.Label(label="Predicted Impact")
 
+    # 输入事件绑定
+    title_input.change(
+        update_button_status,
+        inputs=[title_input, abstract_input],
+        outputs=[validation_status, submit_button]
+    )
+    abstract_input.change(
+        update_button_status,
+        inputs=[title_input, abstract_input],
+        outputs=[validation_status, submit_button]
+    )
+
+    submit_button.click(
+        predict,
+        inputs=[title_input, abstract_input],
+        outputs=output
+    )
+
+    gr.Examples(
+        examples=examples,
+        inputs=[title_input, abstract_input],
+        outputs=[validation_status, output],
+        cache_examples=False
+    )
     gr.Markdown("""
     **Important Notes**  
     - It is intended as a tool for research and educational purposes only.
-    - The web version is a demo for testing purposes. Due to partial environment discrepancies, predictions may have a deviation of up to <0.05.
     - Predicted impact is a probabilistic value generated by the model and does not reflect paper quality or novelty.  
     - The author takes no responsibility for the prediction results.
     - To identify potentially impactful papers, this study uses the sigmoid+MSE approach to optimize NDCG values (over sigmoid+BCE), resulting in predicted values concentrated between 0.1 and 0.9 due to the sigmoid gradient effect.
-    - Generally, we consider a predicted influence score greater than 0.65 to indicate an exceptionally impactful paper.
+    - Generally, it is considered a predicted influence score greater than 0.65 to indicate an exceptionally impactful paper.
     """)
-    submit_button.click(predict, inputs=[title_input, abstract_input], outputs=output)
-
 iface.launch()
 
 
+
+