From 5292f70981f2c1d9a86b2a94c0e7e7a6ca3c3d6c Mon Sep 17 00:00:00 2001
From: Yi Zhu <yizhu59@gmail.com>
Date: Sun, 30 May 2021 19:40:12 -0700
Subject: [PATCH] Fix warmup for i3dslow family (#1666)

* change warmup for i3dslow

* rm load

* try r2+1d
---
 .../i3d_slow_resnet101_f16s4_kinetics400.yaml  | 10 +++++-----
 .../i3d_slow_resnet101_f16s4_kinetics700.yaml  | 10 +++++-----
 .../i3d_slow_resnet101_f32s2_kinetics400.yaml  | 10 +++++-----
 .../i3d_slow_resnet101_f8s8_kinetics400.yaml   | 10 +++++-----
 .../i3d_slow_resnet50_f16s4_kinetics400.yaml   | 10 +++++-----
 .../i3d_slow_resnet50_f32s2_custom.yaml        | 10 +++++-----
 .../i3d_slow_resnet50_f32s2_kinetics400.yaml   | 10 +++++-----
 .../i3d_slow_resnet50_f8s8_kinetics400.yaml    |  8 ++++----
 .../r2plus1d_v1_resnet50_kinetics400.yaml      | 18 +++++++++---------
 .../r2plus1d_v2_resnet152_kinetics400.yaml     | 10 +++++-----
 10 files changed, 53 insertions(+), 53 deletions(-)

diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics400.yaml
index f79bbd11c4..e80387a908 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet101_f16s4_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics700.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics700.yaml
index 40b5188d9e..ddcfb968db 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics700.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet101_f16s4_kinetics700.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet101_f16s4_kinetics700/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet101_f32s2_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet101_f32s2_kinetics400.yaml
index 93dbcc2608..2dfbc85a41 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet101_f32s2_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet101_f32s2_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet101_f32s2_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet101_f8s8_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet101_f8s8_kinetics400.yaml
index e1ee0058ef..7d6cfdad67 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet101_f8s8_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet101_f8s8_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet101_f8s8_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet50_f16s4_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet50_f16s4_kinetics400.yaml
index fb7cad37e6..8b144830dd 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet50_f16s4_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet50_f16s4_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet50_f16s4_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_custom.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_custom.yaml
index e7833e8681..26ee3ee62d 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_custom.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_custom.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet50_f32s2_custom/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_kinetics400.yaml
index 89aa006bb4..96f6d0ad77 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet50_f32s2_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,4 +52,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/i3d_slow_resnet50_f32s2_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/i3d_slow_resnet50_f8s8_kinetics400.yaml b/scripts/action-recognition/configuration/i3d_slow_resnet50_f8s8_kinetics400.yaml
index ee3ac4f606..0471ec0e30 100644
--- a/scripts/action-recognition/configuration/i3d_slow_resnet50_f8s8_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/i3d_slow_resnet50_f8s8_kinetics400.yaml
@@ -16,9 +16,9 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.01
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
@@ -48,9 +48,9 @@ CONFIG:
     PRETRAINED: False
 
   LOG:
-    BASE_PATH: './logs/i3d_slow_resnet50_f8s8_kinetics400_warmup'
+    BASE_PATH: './logs/i3d_slow_resnet50_f8s8_kinetics400'
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
-    EVAL_DIR: './logs/i3d_slow_resnet50_f8s8_kinetics400_warmup/eval'
+    EVAL_DIR: './logs/i3d_slow_resnet50_f8s8_kinetics400/eval'
     DISPLAY_FREQ: 50
     SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/r2plus1d_v1_resnet50_kinetics400.yaml b/scripts/action-recognition/configuration/r2plus1d_v1_resnet50_kinetics400.yaml
index 60683a841f..5fc130e9bd 100644
--- a/scripts/action-recognition/configuration/r2plus1d_v1_resnet50_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/r2plus1d_v1_resnet50_kinetics400.yaml
@@ -5,8 +5,8 @@ DDP_CONFIG:
   WORLD_RANK: 0
   GPU_WORLD_SIZE: 8
   GPU_WORLD_RANK: 0
-  DIST_URL: 'tcp://172.31.72.195:23456'
-  WOLRD_URLS: ['172.31.72.195']
+  DIST_URL: 'tcp://172.31.69.242:23456'
+  WOLRD_URLS: ['172.31.69.242']
   AUTO_RANK_MATCH: True
   DIST_BACKEND: 'nccl'
   GPU: 0
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.001
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -53,8 +52,9 @@ CONFIG:
     PRETRAINED: False
 
   LOG:
-    BASE_PATH: './logs/r2plus1d_v1_resnet50_kinetics400'
+    BASE_PATH: './logs/r2plus1d_v1_resnet50_kinetics400_05302021'
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
-    EVAL_DIR: './logs/r2plus1d_v1_resnet50_kinetics400/eval'
-    SAVE_FREQ: 2
+    EVAL_DIR: './logs/r2plus1d_v1_resnet50_kinetics400_05302021/eval'
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5
diff --git a/scripts/action-recognition/configuration/r2plus1d_v2_resnet152_kinetics400.yaml b/scripts/action-recognition/configuration/r2plus1d_v2_resnet152_kinetics400.yaml
index 88468eba2e..ca6c1f0fb0 100644
--- a/scripts/action-recognition/configuration/r2plus1d_v2_resnet152_kinetics400.yaml
+++ b/scripts/action-recognition/configuration/r2plus1d_v2_resnet152_kinetics400.yaml
@@ -16,17 +16,16 @@ CONFIG:
   TRAIN:
     EPOCH_NUM: 196
     BATCH_SIZE: 8
-    LR: 0.001
+    LR: 0.1
     MOMENTUM: 0.9
-    W_DECAY: 1e-5
+    W_DECAY: 1e-4
     USE_WARMUP: True
     WARMUP_EPOCHS: 34
     LR_POLICY: 'Cosine'
-    WARMUP_END_LR: 0.1
     RESUME_EPOCH: -1
 
   VAL:
-    FREQ: 2
+    FREQ: 5
     BATCH_SIZE: 8
 
   DATA:
@@ -57,4 +56,5 @@ CONFIG:
     LOG_DIR: 'tb_log'
     SAVE_DIR: 'checkpoints'
     EVAL_DIR: './logs/r2plus1d_v2_resnet152_kinetics400/eval'
-    SAVE_FREQ: 2
+    DISPLAY_FREQ: 50
+    SAVE_FREQ: 5