Team NYCU at Defactify4: Robust Detection and Source Identification of AI-Generated Images Using CNN and CLIP-Based Models

ArXiv.org · 2025-03-13

With the rapid advancement of generative AI, AI-generated images have become increasingly realistic, raising concerns about creativity, misinformation, and content authenticity. Detecting such images and identifying their source models has become a critical challenge in ensuring the integrity of digital media. This paper tackles the detection of AI-generated images and identifying their source models using CNN and CLIP-ViT classifiers. For the CNN-based classifier, we leverage EfficientNet-B0 as the backbone and feed with RGB channels, frequency features, and reconstruction errors, while for CLIP-ViT, we adopt a pretrained CLIP image encoder to extract image features and SVM to perform classification. Evaluated on the Defactify 4 dataset, our methods demonstrate strong performance in both tasks, with CLIP-ViT showing superior robustness to image perturbations. Compared to baselines like AEROBLADE and OCC-CLIP, our approach achieves competitive results. Notably, our method ranked Top-3 overall in the Defactify 4 competition, highlighting its effectiveness and generalizability. All of our implementations can be found in https://github.com/uuugaga/Defactify_4

3006eP The genetic aberrations of advanced thyroid cancers in Taiwan: Taiwan cooperative oncology group T2322 study

Annals of Oncology · 2025-09-01

Deep Neural Network Based Microwave GaAs pHEMT Model

2025-05-04

The behavioral model of the gallium arsenide (GaAs) pseudomorphic high electron mobility transistor (pHEMT) established with the deep learning technique is presented in this paper. Backpropagation and adaptive moment estimation (Adam) optimization techniques are utilized to develop a deep neural network (DNN) model. The ReLU activation function is used for neurons of hidden layers and the vanishing gradient problem can be avoided. The characteristics of this transistor are measured by a vector network analyzer (VNA) to obtain <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$S$</tex>-parameters up to 40 GHz. This measurement system is calibrated using the short-open-load-thru (SOLT) method. These measured <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$S$</tex>-parameters are de-embedded with the industry standard two-step method to obtain GaAs pHEMT data without the parasitic effects of pads and interconnect lines. The data are separated into training and testing datasets to obtain and validate this model. The input layer contains three neurons with the frequency, gate-source bias voltage <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{V}_{\text{gs}}$</tex> and drain-source bias voltage <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{V}_{\text{ds}}$</tex> and the output layer consists of eight neurons with real and imaginary parts of the de-embedded <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$S$</tex>-parameters. 1000 epochs are performed for the training procedure to achieve a mean square error (MSE) lower than the required value. The dropout technique is utilized to avoid the over-fitting problem. TensorFlow and Keras are used to build this DNN. The measured and predicted data obtained from this DNN are in a good agreement. Compared with conventional shallow neural networks, accurate results can be achieved with a smaller number of neurons. This novel GaAs pHEMT behavioral model can be applied to microwave circuit design efficiently and accurately.

Adipocyte-specific Nrf2 deletion negates nitro-oleic acid benefits on glucose tolerance in diet-induced obesity

Nitric Oxide · 2024-06-13 · 2 citations

Obesity is commonly linked with adipose tissue (AT) dysfunction, setting off inflammation and oxidative stress, both key contributors to the cardiometabolic complications associated with obesity. To improve metabolic and cardiovascular health, countering these inflammatory and oxidative signaling processes is crucial. Offering potential in this context, the activation of nuclear factor erythroid 2-related factor 2 (Nrf2) by nitro-fatty acids (NO2-FA) promote diverse anti-inflammatory signaling and counteract oxidative stress. Additionally, we previously highlighted that nitro-oleic acid (NO2-OA) preferentially accumulates in white adipose tissue (AT) and provides protection against already established high fat diet (HFD)-mediated impaired glucose tolerance. The precise mechanism accounting for these protective effects remained largely unexplored until now. Herein, we reveal that protective effects of improved glucose tolerance by NO2-OA is absent when Nrf2 is specifically ablated in adipocytes (ANKO mice). NO2-OA treatment did not alter body weight between ANKO and littermate controls (Nrf2fl/fl) mice on both the HFD and low-fat diet (LFD). As expected, at day 76 (before NO2-OA treatment) and notably at day 125 (daily treatment of 15 mg/kg NO2-OA for 48 days), both HFD-fed Nrf2fl/fl and ANKO mice exhibited increased fat mass and reduced lean mass compared to LFD controls. However, throughout the NO2-OA treatment, no distinction was observed between Nrf2fl/fl and ANKO in the HFD-fed mice as well as in the Nrf2fl/fl mice fed a LFD. Glucose tolerance tests revealed impaired glucose tolerance in HFD-fed Nrf2fl/fl and ANKO compared to LFD-fed Nrf2fl/fl mice. Notably, NO2-OA treatment improved glucose tolerance in HFD-fed Nrf2fl/fl but did not yield the same improvement in ANKO mice at days 15, 30, and 55 of treatment. Unraveling the pathways linked to NO2-OA's protective effects in obesity-mediated impairment in glucose tolerance is pivotal within the realm of precision medicine, crucially propelling future applications and refining novel drug-based strategies.

Pain, function and peritendinous effusion improvement after dry needling in patients with long head of biceps brachii tendinopathy: a single-blind randomized clinical trial

Annals of Medicine · 2024-08-14 · 4 citations

INTRODUCTION: Long head of biceps brachii tendinopathy, a frequent source of anterior shoulder pain, may lead to discomfort and diminished function. The objective of this study is to assess the efficacy of dry needling and transcutaneous electrical nerve stimulation in these patients. PATIENTS AND METHODS: Thirty patients were randomized into dry needling and transcutaneous electrical nerve stimulation groups and assessed before treatment, 8 and 15 days after treatment using a visual analogue scale, shoulder pain and disability index, pressure pain threshold, tissue hardness, and biceps peritendinous effusion. RESULTS: < 0.01). CONCLUSIONS: Dry needling showed non-inferior results to transcutaneous electrical nerve stimulation in reducing pain and disability and demonstrated even superior results in reducing biceps peritendinous effusion (see Graphical Abstract). TRIAL REGISTRATION: The Institutional Review Board of the China Medical University Hospital (CMUH107-REC2-101) approved this study, and it was registered with Identifier NCT03639454 on ClinicalTrials.gov.

Comment to: Efficacy and safety of mesh non‑fixation in patients undergoing laparo‑endoscopic repair of groin hernia

Hernia · 2024-01-12

Comment to: Can ventral transabdominal preperitoneal achieve favorable outcomes in minimally invasive ventral hernia repair?

Hernia · 2023-09-15

(Invited) Some Perspectives on Thermodynamics, Kinetics, Mechanics and Phase Transformations of Ionic and Mixed Conducting Ceramics

ECS Meeting Abstracts · 2023-12-22

A perhaps oversimplified description of Professor Virkar's research of the last 45 years is thermodynamics, kinetics, (fracture) mechanics--which is thermodynamics at 0 Kelvin, and phase transformations of ceramic materials. I will reflect on some of his significant contributions in these areas, and echo them with a few recent research notes of my own.

Prognosis of Paraplegia after Receiving Transarterial Chemoembolization for Recurrent Hepatocellular Carcinoma: A Case Report

Rehabilitation practice and science · 2023-01-01 · 1 citations

Transarterial chemoembolization (TACE) is the standard treatment for patients with Barcelona clinic liver cancer staging system stage B hepatocellular carcinoma. However, various TACE complications have been reported, and spinal infarction is one of the rarest. To date, only six specific case reports written in English have been published. We report a rare case of TACE-related spinal cord infarction and compare this case with cases in other reports and literature reviews. The patient experienced a sudden onset of paraplegia, sensory impairment below the T8 dermatome and had difficultly voiding immediately after TACE. The T2WI and DWI found hyperintensity 3.5 hours after the onset of symptoms. The patient received high-dose cerebrolysin, hyperbaric oxygen therapy, and traditional rehabilitation. However, 10 months after TACE, motor symptoms were not alleviated. Meanwhile, sensory symptoms were gradually alleviated. The impact of this rare complication to personal life, family, and social participation is overwhelming. Therefore, we should direct our attention to this rare complication and its prognosis.

Comment to: should routine surgical wound drainage after ventral hernia repair be avoided?

Hernia · 2023-09-16