train.sh 2.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687
  1. # Dataset setting
  2. DATASET="coco"
  3. DATA_ROOT="/data/datasets/"
  4. # DATA_ROOT="/Users/liuhaoran/Desktop/python_work/object-detection/dataset/"
  5. # MODEL setting
  6. MODEL="yolov8_n"
  7. IMAGE_SIZE=640
  8. RESUME="None"
  9. if [[ $MODEL == *"yolov8"* ]]; then
  10. # Epoch setting
  11. BATCH_SIZE=128
  12. MAX_EPOCH=500
  13. WP_EPOCH=3
  14. EVAL_EPOCH=10
  15. NO_AUG_EPOCH=20
  16. elif [[ $MODEL == *"yolox"* ]]; then
  17. # Epoch setting
  18. BATCH_SIZE=64
  19. MAX_EPOCH=300
  20. WP_EPOCH=3
  21. EVAL_EPOCH=10
  22. NO_AUG_EPOCH=15
  23. elif [[ $MODEL == *"yolov7"* ]]; then
  24. # Epoch setting
  25. BATCH_SIZE=128
  26. MAX_EPOCH=300
  27. WP_EPOCH=3
  28. EVAL_EPOCH=10
  29. NO_AUG_EPOCH=20
  30. elif [[ $MODEL == *"yolov5"* || $MODEL == *"yolov4"* || $MODEL == *"yolov3"* ]]; then
  31. # Epoch setting
  32. BATCH_SIZE=128
  33. MAX_EPOCH=300
  34. WP_EPOCH=3
  35. EVAL_EPOCH=10
  36. NO_AUG_EPOCH=15
  37. else
  38. # Epoch setting
  39. BATCH_SIZE=128
  40. MAX_EPOCH=150
  41. WP_EPOCH=3
  42. EVAL_EPOCH=10
  43. NO_AUG_EPOCH=0
  44. fi
  45. # -------------------------- Train Pipeline --------------------------
  46. WORLD_SIZE=1
  47. if [ $WORLD_SIZE == 1 ]; then
  48. python train.py \
  49. --cuda \
  50. --dataset ${DATASET} \
  51. --root ${DATA_ROOT} \
  52. --model ${MODEL} \
  53. --batch_size ${BATCH_SIZE} \
  54. --img_size ${IMAGE_SIZE} \
  55. --wp_epoch ${WP_EPOCH} \
  56. --max_epoch ${MAX_EPOCH} \
  57. --eval_epoch ${EVAL_EPOCH} \
  58. --no_aug_epoch ${NO_AUG_EPOCH} \
  59. --resume ${RESUME} \
  60. --ema \
  61. --fp16 \
  62. --multi_scale
  63. elif [[ $WORLD_SIZE -gt 1 && $WORLD_SIZE -le 8 ]]; then
  64. python -m torch.distributed.run --nproc_per_node=8 train.py \
  65. --cuda \
  66. -dist \
  67. --dataset ${DATASET} \
  68. --root ${DATA_ROOT} \
  69. --model ${MODEL} \
  70. --batch_size ${BATCH_SIZE} \
  71. --img_size ${IMAGE_SIZE} \
  72. --wp_epoch ${WP_EPOCH} \
  73. --max_epoch ${MAX_EPOCH} \
  74. --eval_epoch ${EVAL_EPOCH} \
  75. --no_aug_epoch ${NO_AUG_EPOCH} \
  76. --resume ${RESUME} \
  77. --ema \
  78. --fp16 \
  79. --multi_scale \
  80. --sybn
  81. else
  82. echo "The WORLD_SIZE is set to a value greater than 8, indicating the use of multi-machine \
  83. multi-card training mode, which is currently unsupported."
  84. exit 1
  85. fi