mirrors
/
fast-reid
mirror of https://github.com/JDAI-CV/fast-reid.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Update bagtricks_R50-ibn.yml 

The config that obtains the best performance on VeRi-Wild. The model is trained on 4 P40 GPUs
pull/395/head
Xinchen Liu 2021-01-28 11:15:57 +08:00 committed by GitHub
parent 1ed1a13eed
commit 885dc96608
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 13 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
configs/VERIWild/bagtricks_R50-ibn.yml
View File

@ -3,13 +3,18 @@ _BASE_: ../Base-bagtricks.yml
INPUT:
SIZE_TRAIN: [256, 256]
SIZE_TEST: [256, 256]
CJ:
ENABLED: False
MODEL:
BACKBONE:
WITH_IBN: True
WITH_NL: False
HEADS:
NAME: EmbeddingHead
POOL_LAYER: gempool
LOSSES:
NAME: ("CrossEntropyLoss", "TripletLoss",)
TRI:
HARD_MINING: False
MARGIN: 0.0
@ -19,15 +24,18 @@ DATASETS:
TESTS: ("SmallVeRiWild", "MediumVeRiWild", "LargeVeRiWild",)
SOLVER:
IMS_PER_BATCH: 128
MAX_ITER: 60
STEPS: [30, 50]
WARMUP_ITERS: 2000
BASE_LR: 0.00035
BIAS_LR_FACTOR: 1.0
ETA_MIN_LR: 3e-07
IMS_PER_BATCH: 512 # 512 For 4 GPUs, 128 per GPU
MAX_EPOCH: 120
STEPS: [30, 70, 90]
WARMUP_ITERS: 5000
CHECKPOINT_PERIOD: 20
TEST:
EVAL_PERIOD: 20
EVAL_PERIOD: 10
IMS_PER_BATCH: 128
OUTPUT_DIR: logs/veriwild/bagtricks_R50-ibn_4gpu