dvlab-research/MSAD dvlab-research/MSAD

  • Stars
    star
    127
  • Rank 282,790 (Top 6 %)
  • Language
    Python
  • Created over 3 years ago
  • Updated over 3 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
dvlab-research/MSAD
github

dvlab-research

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Multi-Scale Aligned Distillation for Low-Resolution Detection (CVPR2021)

MSAD

Multi-Scale Aligned Distillation for Low-Resolution Detection

Lu Qi*, Jason Kuen*, Jiuxiang Gu, Zhe Lin, Yi Wang, Yukang Chen, Yanwei Li, Jiaya Jia


This project provides an implementation for the CVPR 2021 paper "Multi-Scale Aligned Distillation for Low-Resolution Detection" based on Detectron2. MSAD targets to detect objects using low-resolution instead of high-resolution image. MSAD could obtain comparable performance in high-resolution image size. Our paper use Slimmable Neural Networks as our pretrained weight.

Installation

This project is based on Detectron2, which can be constructed as follows.

  • Install Detectron2 following the instructions. We are noting that our code is checked in detectron2 V0.2.1 (commit version: be792b959bca9af0aacfa04799537856c7a92802) and pytorch 1.4.
  • Setup the dataset following the structure.
  • Copy this project to /path/to/detectron2/projects/MSAD
  • Download the slimmable networks in the github. The slimmable resnet50 pretrained weight link is here.
  • Set the "find_unused_parameters=True" in distributed training of your own detectron2. You could modify it in detectron2/engine/defaults.py.

Pretrained Weight

  • Move the pretrained weight to your target path
  • Modify the weight path in configs/Base-SLRESNET-FCOS.yaml

Teacher Training

To train teacher model with 8 GPUs, run:

cd /path/to/detectron2
python3 projects/MSAD/train_net_T.py --config-file <projects/MSAD/configs/config.yaml> --num-gpus 8

For example, to launch MSAD teacher training (1x schedule) with Slimmable-ResNet-50 backbone in 0.25 width on 8 GPUs and save the model in the path "/data/SLR025-50-T". one should execute:

cd /path/to/detectron2
python3 projects/MSAD/train_net_T.py --config-file projects/MSAD/configs/SLR025-50-T.yaml --num-gpus 8 OUTPUT_DIR /data/SLR025-50-T

Student Training

To train student model with 8 GPUs, run:

cd /path/to/detectron2
python3 projects/MSAD/train_net_S.py --config-file <projects/MSAD/configs/config.yaml> --num-gpus 8

For example, to launch MSAD student training (1x schedule) with Slimmable-ResNet-50 backbone in 0.25 width on 8 GPUs and save the model in the path "/data/SLR025-50-S". We assume the teacher weight is saved in the path "/data/SLR025-50-T/model_final.pth" one should execute:

cd /path/to/detectron2
python3 projects/MSAD/train_net_S.py --config-file projects/MSAD/configs/MSAD-R50-S025-1x.yaml --num-gpus 8 MODEL.WEIGHTS /data/SLR025-50-T/model_final.pth OUTPUT_DIR MSAD-R50-S025-1x

Evaluation

To evaluate a teacher or student pre-trained model with 8 GPUs, run:

cd /path/to/detectron2
python3 projects/MSAD/train_net_T.py --config-file <config.yaml> --num-gpus 8 --eval-only MODEL.WEIGHTS model_checkpoint

or

cd /path/to/detectron2
python3 projects/MSAD/train_net_S.py --config-file <config.yaml> --num-gpus 8 --eval-only MODEL.WEIGHTS model_checkpoint

Results

We provide the results on COCO val set with pretrained models. In the following table, we define the backbone FLOPs as capacity. For brevity, we regard the FLOPs of Slimmable Resnet50 in width 1.0 and high resolution input (800,1333) as 1x. The metrics are reported in old-version detectron2. The new-version detectron will report higher loss value but it does not affect the final result.

Method Backbone Capacity Sched Width Role Resolution BoxAP download
FCOS Slimmable-R50 1.25x 1x 1.00 Teacher H & L 42.8 model | metrics
FCOS Slimmable-R50 0.25x 1x 1.00 Student L 39.9 model | metrics
FCOS Slimmable-R50 0.70x 1x 0.75 Teacher H & L 41.2 model | metrics
FCOS Slimmable-R50 0.14x 1x 0.75 Student L 38.8 model | metrics
FCOS Slimmable-R50 0.31x 1x 0.50 Teacher H & L 38.4 model | metrics
FCOS Slimmable-R50 0.06x 1x 0.50 Student L 35.7 model | metrics
FCOS Slimmable-R50 0.08x 1x 0.25 Teacher H & L 33.2 model | metrics
FCOS Slimmable-R50 0.02x 1x 0.25 Student L 30.3 model | metrics

Citing MSAD

Consider cite MSAD in your publications if it helps your research.

@article{qi2021msad,
  title={Multi-Scale Aligned Distillation for Low-Resolution Detection},
  author={Lu Qi, Jason Kuen, Jiuxiang Gu, Zhe Lin, Yi Wang, Yukang Chen, Yanwei Li, Jiaya Jia},
  journal={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2021}
}

More Repositories

1

MGM

Official repo for "Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models"
Python
3,111
star
2

LongLoRA

Code and documents of LongLoRA and LongAlpaca (ICLR 2024 Oral)
Python
2,563
star
3

LISA

Project Page for "LISA: Reasoning Segmentation via Large Language Model"
Python
1,680
star
4

VoxelNeXt

VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023)
Python
692
star
5

LLaMA-VID

LLaMA-VID: An Image is Worth 2 Tokens in Large Language Models (ECCV 2024)
Python
656
star
6

DeepUPE

Underexposed Photo Enhancement Using Deep Illumination Estimation
Python
566
star
7

3D-Box-Segment-Anything

We extend Segment Anything to 3D perception by combining it with VoxelNeXt.
Jupyter Notebook
524
star
8

LLMGA

This project is the official implementation of 'LLMGA: Multimodal Large Language Model based Generation Assistant', ECCV2024
Python
424
star
9

ControlNeXt

Controllable video and image Generation, SVD, Animate Anyone, ControlNet, LoRA
Python
417
star
10

PanopticFCN

Fully Convolutional Networks for Panoptic Segmentation (CVPR2021 Oral)
Python
391
star
11

PointGroup

PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation
Python
376
star
12

3DSSD

3DSSD: Point-based 3D Single Stage Object Detector (CVPR 2020)
Python
375
star
13

Video-P2P

Video-P2P: Video Editing with Cross-attention Control
Python
365
star
14

FocalsConv

Focal Sparse Convolutional Networks for 3D Object Detection (CVPR 2022, Oral)
Python
364
star
15

Stratified-Transformer

Stratified Transformer for 3D Point Cloud Segmentation (CVPR 2022)
Python
362
star
16

DSGN

DSGN: Deep Stereo Geometry Network for 3D Object Detection (CVPR 2020)
Python
324
star
17

PFENet

PFENet: Prior Guided Feature Enrichment Network for Few-shot Segmentation (TPAMI).
Python
307
star
18

SphereFormer

The official implementation for "Spherical Transformer for LiDAR-based 3D Recognition" (CVPR 2023).
Python
300
star
19

GridMask

Python
281
star
20

ReviewKD

Distilling Knowledge via Knowledge Review, CVPR 2021
Python
249
star
21

Parametric-Contrastive-Learning

Parametric Contrastive Learning (ICCV2021) & GPaCo (TPAMI 2023)
Python
237
star
22

Step-DPO

Implementation for "Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs"
Python
232
star
23

Simple-SR

Include MuCAN, LAPAR, etc.
Python
224
star
24

UVTR

Unifying Voxel-based Representation with Transformer for 3D Object Detection (NeurIPS 2022)
Python
224
star
25

Facelet_Bank

Facelet-Bank for Fast Portrait Manipulation (pytorch)
Python
208
star
26

SA-AutoAug

Scale-aware Automatic Augmentation for Object Detection (CVPR 2021)
Python
196
star
27

LargeKernel3D

LargeKernel3D: Scaling up Kernels in 3D Sparse CNNs (CVPR 2023)
Python
189
star
28

SNR-Aware-Low-Light-Enhance

This is the official implementation for the paper "SNR-aware low-light image enhancement" in CVPR2022
Python
160
star
29

MASA-SR

MASA-SR: Matching Acceleration and Spatial Adaptation for Reference-Based Image Super-Resolution (CVPR2021)
Python
158
star
30

ECCV22-P3AFormer-Tracking-Objects-as-Pixel-wise-Distributions

The official code for our ECCV22 oral paper: tracking objects as pixel-wise distributions.
Python
158
star
31

Context-Aware-Consistency

Semi-supervised Semantic Segmentation with Directional Context-aware Consistency (CVPR 2021)
Python
155
star
32

SparseTransformer

A fast and memory-efficient libarary for sparse transformer with varying token numbers (e.g., 3D point cloud).
Python
153
star
33

spconv-plus

Python
152
star
34

EfficientNeRF

The official code for "Efficient Neural Radiance Fields" in CVPR2022.
Python
152
star
35

MiSLAS

Improving Calibration for Long-Tailed Recognition (CVPR2021)
Python
144
star
36

RIVAL

[NeurIPS 2023 Spotlight] Real-World Image Variation by Aligning Diffusion Inversion Chain
Python
143
star
37

MOOD

Official PyTorch implementation of MOOD series: (1) MOODv1: Rethinking Out-of-distributionDetection: Masked Image Modeling Is All You Need. (2) MOODv2: Masked Image Modeling for Out-of-Distribution Detection.
Python
133
star
38

outpainting_srn

Wide-Context Semantic Image Extrapolation, CVPR2019
Python
131
star
39

DeepVision3D

DeepVision3D is an open source toolbox for point-cloud understanding.
Python
119
star
40

Ref-NPR

[CVPR 2023] Ref-NPR: Reference-Based Non-PhotoRealistic Radiance Fields
Python
119
star
41

VFIformer

Video Frame Interpolation with Transformer (CVPR2022)
Python
112
star
42

Prompt-Highlighter

[CVPR 2024] Prompt Highlighter: Interactive Control for Multi-Modal LLMs
Python
112
star
43

VFF

Voxel Field Fusion for 3D Object Detection (CVPR2022)
Python
95
star
44

SMR

Self-Supervised 3D Mesh Reconstruction from Single Images (CVPR2021)
Python
91
star
45

SCGAN

The implementation of 'Image synthesis via semantic composition', ICCV2021.
Python
81
star
46

Imbalanced-Learning

Imbalanced learning tool for imbalanced recognition and segmentation
Python
79
star
47

JigsawClustering

This is the code for CVPR 2021 oral paper: Jigsaw Clustering for Unsupervised Visual Representation Learning
Python
78
star
48

AttenNorm

Attentive Normalization for Conditional Image Generation
Python
71
star
49

GFS-Seg

The official implementation of Generalized Few-shot Semantic Segmentation (CVPR 2022)
Python
63
star
50

Mask-Attention-Free-Transformer

Official Implementation for "Mask-Attention-Free Transformer for 3D Instance Segmentation"
Python
59
star
51

MoTCoder

This is the official code repository of MoTCoder: Elevating Large Language Models with Modular of Thought for Challenging Programming Tasks.
Python
58
star
52

SDSD

Seeing Dynamic Scene in the Dark: High-Quality Video Dataset with Mechatronic Alignment (ICCV2021)
Python
48
star
53

GroupContrast

[CVPR 2024] GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding
42
star
54

ProposeReduce

Video Instance Segmentation with a Propose-Reduce Paradigm (ICCV 2021)
Python
41
star
55

Robust-Semantic-Segmentation

Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation （ICCV2021）
Python
40
star
56

Mr-Ben

This is the repo for our paper "Mr-Ben: A Comprehensive Meta-Reasoning Benchmark for Large Language Models"
Python
38
star
57

BAL

BAL: Balancing Diversity and Novelty for Active Learning - Official Pytorch Implementation
Python
38
star
58

TriVol

The official code of TriVol in CVPR-2023
Python
37
star
59

MR-GSM8K

Challenge LLMs to Reason About Reasoning: A Benchmark to Unveil Cognitive Depth in LLMs
Python
37
star
60

DecoupleNet

Official implementation for our ECCV 2022 paper "DecoupleNet: Decoupled Network for Domain Adaptive Semantic Segmentation"
Python
36
star
61

Dsig

Deep Structured Instance Graph for Distilling Object Detectors (ICCV 2021)
Python
35
star
62

LBGAT

Learnable Boundary Guided Adversarial Training (ICCV2021)
Python
33
star
63

Q-LLM

This is the official repo of "QuickLLaMA: Query-aware Inference Acceleration for Large Language Models"
Python
31
star
64

AGSS-VOS

AGSS-VOS: Attention Guided Single-Shot Video Object Segmentation
Python
20
star
65

MAT

MAT: Mask-Aware Transformer for Large Hole Image Inpainting
Python
16
star
66

MSN

Memory Selection Network for Video Propagation (ECCV 2020)
Python
6
star
67

APD

Python
5
star
68

Point2Pix

The official code of Point2pix in CVPR-2023
2
star
69

TagCLIP

Python
2
star