# IHC paper
###### tags: `IHC_paper`

# Model dataset
- veri776,veri776_synreal,veriwild,veriwild_synreal
- on TWCC
- 在國網 VeRi_syn 的 image_test_hazy 是 image_test_b0b5
- hazy: b0b5, synreal: b0~b2 (random), syn: b0-b3

# 我分的haze datset

# My question
- how to compare
- train on veri_synreal
- train on wild_synreal
- train on wild_clear
# ImageNet
http://www.image-net.org/download-images
# Things have not run yet
- run b
# Code have not run yet
- RAM VEHICLE RE-IDENTIFICATION
- paper:https://arxiv.org/pdf/1806.09283.pdf
- code : https://arxiv.org/pdf/1806.09283.pdf
- VOC-ReID
- paper: https://arxiv.org/pdf/2004.09164.pdf
- code:https://github.com/Xiangyu-CAS/AICity2020-VOC-ReID
- Baidu-UTS
- paper:
- code : https://github.com/layumi/AICIty-reID-2020
# syn veri_wild dataset
- depth : /mnt/190/a/Cihsiang/ReID/Dataset/VERI_WILD_hazehelp/images_all_depth_outdoor/
- folder name :veriwild_syn_real
- make haze(syn) code:
- /mnt/190/a/Cihsiang/ReID/1.Make_Depth/VeRi_Syn_hazy/dense_haze_OTS_veri_train_synreal.m"

# Code have not run
- DMT
- reuslt path:/work/r09921058/ReID_dehazeREID/ReID_inference/Inference_veri_synreal/

# Hackmd

# old data pretrain on b0_epoch80
- x
* DMT Train on veri_syn_b0
- TEST_SYN_b0:
| map | cmc1 | cmc5 | cmc10 |
|:---- |:---- |:---- |:----- |
| 81.1 | 96.6 | 98.6 | 99.2 |
- TEST_SYN_b1:
| map | cmc1 | cmc5 | cmc10 |
|:---- |:---- |:---- |:----- |
| 69.6 | 92 | 97 | 98.4 |
- TEST_SYN_b2:
| map | cmc1 | cmc5 | cmc10 |
|:---- |:---- |:----- |:----- |
| 59.7 | 88.5 | 94.30 | 96.5 |
- TEST_SYN_b3:
| map | cmc1 | cmc5 | cmc10 |
| ---- |:---- |:---- |:----- |
| 50.5 | 84.3 | 91.5 | 94.8 |
* DMT Train on veri_syn_b1
- TEST_SYN_b1:
| map | cmc1 | cmc5 | cmc10 |
| ---- |:---- | ---- |:----- |
| 78.5 | 96.1 | 98.5 | 99.2 |
* DMT Train on veri_syn_b2
- TEST_SYN_b2:
| map | cmc1 | cmc5 | cmc10 |
| ---- |:---- | ---- |:----- |
| 76.4 | 95.5 | 97.9 | 98.8 |
* DMT Train on veri_syn_b3
- TEST_SYN_b3:
| map | cmc1 | cmc5 | cmc10 |
| --- |:---- |:---- |:----- |
| 73 | 94.7 | 97.6 | 98.7 |
* DMT Train on veri_syn_b4
- TEST_SYN_b3:
| map | cmc1 | cmc5 | cmc10 |
| --- |:---- |:---- |:----- |
| 73 | 94.7 | 97.6 | 98.7 |
* DMT Train on veri_syn_b5
- TEST_SYN_b3:
| map | cmc1 | cmc5 | cmc10 |
| --- |:---- |:---- |:----- |
| 73 | 94.7 | 97.6 | 98.7 |

# Paper survey
## Aicity 2019
- https://openaccess.thecvf.com/content_CVPRW_2019/papers/AI%20City/Naphade_The_2019_AI_City_Challenge_CVPRW_2019_paper.pdf
## Aicity 2020
- 

* Multi-Domain Learning and Identity Mining for Vehicle Re-Identification(CVPR 2020 Workshop)
* 
* resNet pretrain 80
* 100 epoches
* inb_a :https://github.com/XingangPan/IBN-Net

## multi task or part mask
-(no code) https://openaccess.thecvf.com/content_CVPR_2019/papers/He_Part-Regularized_Near-Duplicate_Vehicle_Re-Identification_CVPR_2019_paper.pdf

# RBSM
- 要改./configs (add .yml)

- sh train_veh1m_wild.sh

# How to compare
- loss
- SOTA method

# MSBDN result
- image_query_syn
- ===> Avg. SR SSIM: 0.9281
- Avg. SR PSNR:23.537345 dB
- image_query_synreal
- ===> Avg. SR SSIM: 0.9115
- Avg. SR PSNR:22.640361 dB