일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | |||
5 | 6 | 7 | 8 | 9 | 10 | 11 |
12 | 13 | 14 | 15 | 16 | 17 | 18 |
19 | 20 | 21 | 22 | 23 | 24 | 25 |
26 | 27 | 28 | 29 | 30 | 31 |
- 주식
- 논문리뷰
- 주식 용어정리
- 3C4P
- 모델 구조 변경
- mPLUG-2
- 파라미터 수 확인
- Foundation Transformers
- 특정 layer 추출
- KOSMOS-2
- def validation
- 강화학습
- 특정 layer 동결
- mqtt
- cnn
- MLLM
- 논문 작성
- 파라미터 수
- DeepNet
- 가중치 없이 모델 로드
- Video Understanding
- 가상환경
- Instruction dataset
- 모델 freeze
- Multimodal Large Language Model
- layer 추출
- def train
- 모델 동결
- 논문 작성 요령
- pretrained model layer
- Today
- Total
목록전체 글 (115)
시작은 미약하였으나 , 그 끝은 창대하리라
궁금증: mPLUG-2 논문을 보니 concat이라는 부분이 있었는데 깃허브 코드를 확인해보았을때는 concat의 부분이 보이지 않음. 찾아보디 Conv3d에서 인자값으로 group이 존재. 1. group 기능 : Grouped Convolution 은 입력 값의 채널들을 여러 개의 그룹으로 나누어 독립적으로 Convolution 연산을 수행하는 방식. ex) in_channels =4, group=2 이면 → 2개씩 묶어서 진행후 마지막 출력을 concat함. (그래야 쪼개진 입력차원에 대한 정보(차원크기)를 복원할 수 있으니, 예를들어 ) 기타 (딥러닝에 사용되는 다양한 Convolution 기법) : https://eehoeskrap.tistory.com/431#8.-grouped-convolut..
논문 링크: https://www.mdpi.com/2504-446X/7/2/114 Semantic Scene Understanding with Large Language Models on Unmanned Aerial Vehicles Unmanned Aerial Vehicles (UAVs) are able to provide instantaneous visual cues and a high-level data throughput that could be further leveraged to address complex tasks, such as semantically rich scene understanding. In this work, we built on the use of Lar www.mdpi.co..
논문링크 : https://arxiv.org/abs/2302.00402 mPLUG-2: A Modularized Multi-modal Foundation Model Across Text, Image and Video Recent years have witnessed a big convergence of language, vision, and multi-modal pretraining. In this work, we present mPLUG-2, a new unified paradigm with modularized design for multi-modal pretraining, which can benefit from modality collaboration whil arxiv.org Published ..
► 논문은 두괄식 작성이다. 문단을 작성할때 항상 하고싶은말(중요문장)을 맨 앞줄에 쓰고 내용을 서술한다. 한국인인 우리들은 이게 참 어색하다.. (외국애들은 중,고등학교, 대학교에서 배운다고 하더라....교수님이그러심.. , 한국의 교육과정은.......아쉽구나(?)) ► 논문 VS 보고서 대학원 1학기 시절 SCI논문을 작성할 때 가장 많이 들은 이야기이다. 교수님은 내가 작성한 논문은 논문이라기 보다는 보고서 느낌이 강하다는 것이다. 내 입장에서는 어떤의미인지 와 닫지가 않아 국어사전에 정의도 찾아보고, 유튜브, 구글링 등등 다양하게 시도를 해보았다. 그래서 내가 내린결론은 다음과 같다. 보고서 논문 • 누군가에게 보고를 하는 느낌 • 의견을 제시하는 느낌. • 현황및 결과를 나열 하는 느낌 • 누..
논문 링크 : https://arxiv.org/abs/2305.06355 VideoChat: Chat-Centric Video Understanding In this paper, we initiate an attempt of developing an end-to-end chat-centric video understanding system, coined as VideoChat. It integrates video foundation models and large language models via a learnable neural interface, excelling in spatiotemporal re arxiv.org Published : 2021.03.24 (arxiv - 24.01.14 기준) C..
논문 링크 : https://arxiv.org/abs/2103.13313 In-flight positional and energy use data set of a DJI Matrice 100 quadcopter for small package delivery We autonomously direct a small quadcopter package delivery Uncrewed Aerial Vehicle (UAV) or "drone" to take off, fly a specified route, and land for a total of 209 flights while varying a set of operational parameters. The vehicle was equipped with onbo..
❏ Evaluation code ✓ train, validation 을 통한 early stopping을 수행하였으면 나온 모델을 통해 실제 예측을 수행해야하는 마지막 단계에 도달한다. 이를 위해 아래와 같이 코드를 적용하면 된다. ### Evaluation code ### ### ### # 예측값 저장하는 빈 리스트 all_predictions = [] # 모델을 평가모드로 설정. model.eval() for inputs in test_dataloader: with torch.no_grad(): inputs=inputs[0].float() x = inputs.to(device) outputs = model(x) max_values, max_indices = torch.max(outputs, 1) # ..
용도: visual studio 로 c언어를 작성해본적이 있다면 빨간점을 클릭하면 그 부분에 대해서 자세히 들어가며 디버깅을 할 수 있다. 사용 : Visual studio code 의 ipynb 확장자로 코드를 작성후 디버깅하고 싶은 부분에 import pdb;pdb.set_trace() 를 추가하면 새로운 창이 나온다. 그 줄을 기준으로 이전의 변수, 값 들은 다 할당을 받은 상태임. visual studio 의 디버깅 기능과 동일하다 보면 됨. 사용법 : 창이 나오면 n (next), q(quit), s(step) 등의 명령어를 통해 진행할 수 있다. n를 사용하면 다음줄로 넘어가고, s로 하면 그 단계에서 있는 내부 함수등으로 들어가고 q를 하면 종료 된다. 이것외에도 명령어가 있는데 공식 문서를..