add RegNet

[kaijieshi7]

Add RegNet

• build model
• update init.py in models
• convert pretrained weight
• inference test on imagenet and update model_zoo
• update docs
• update readme
• update changelog
• pytorch speed comparison

[kaijieshi7]

OneFlow regnet_y_400mf time: 55.9ms (= 5588.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_400mf time: 63.7ms (= 6373.9ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.14 (= 63.7ms / 55.9ms)

OneFlow regnet_y_800mf time: 48.1ms (= 4810.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_800mf time: 40.5ms (= 4046.9ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.84 (= 40.5ms / 48.1ms)

OneFlow regnet_y_1_6gf time: 79.5ms (= 7949.1ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_1_6gf time: 70.8ms (= 7083.4ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.89 (= 70.8ms / 79.5ms)

OneFlow regnet_y_3_2gf time: 82.9ms (= 8291.1ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_3_2gf time: 79.5ms (= 7945.0ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.96 (= 79.5ms / 82.9ms)

OneFlow regnet_y_8gf time: 135.6ms (= 13556.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_8gf time: 136.5ms (= 13647.1ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.01 (= 136.5ms / 135.6ms)

OneFlow regnet_y_16gf time: 203.7ms (= 20367.8ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_16gf time: 204.1ms (= 20412.9ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.00 (= 204.1ms / 203.7ms)

OneFlow regnet_y_32gf time: 339.5ms (= 33946.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_y_32gf time: 341.6ms (= 34159.0ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.01 (= 341.6ms / 339.5ms)

OneFlow regnet_x_400mf time: 43.2ms (= 4320.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_400mf time: 50.7ms (= 5073.4ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.17 (= 50.7ms / 43.2ms)

OneFlow regnet_x_800mf time: 45.4ms (= 4539.3ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_800mf time: 42.7ms (= 4270.2ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.94 (= 42.7ms / 45.4ms)

OneFlow regnet_x_1_6gf time: 49.7ms (= 4971.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_1_6gf time: 47.2ms (= 4723.0ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.95 (= 47.2ms / 49.7ms)

OneFlow regnet_x_3_2gf time: 79.5ms (= 7945.4ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_3_2gf time: 67.9ms (= 6788.5ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 0.85 (= 67.9ms / 79.5ms)

OneFlow regnet_x_8gf time: 100.5ms (= 10050.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_8gf time: 101.1ms (= 10109.9ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.01 (= 101.1ms / 100.5ms)

OneFlow regnet_x_16gf time: 176.5ms (= 17653.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_16gf time: 177.4ms (= 17740.6ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.00 (= 177.4ms / 176.5ms)

OneFlow regnet_x_32gf time: 335.5ms (= 33548.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch regnet_x_32gf time: 334.9ms (= 33491.0ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.00 (= 334.9ms / 335.5ms)

OneFlow regnet_y_400mf time: 12.4ms (= 1241.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_400mf time: 12.7ms (= 1269.4ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
✔️ Relative speed: 1.02 (= 12.7ms / 12.4ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_800mf time: 12.7ms (= 1266.3ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_800mf time: 9.7ms (= 972.4ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.77 (= 9.7ms / 12.7ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_1_6gf time: 19.9ms (= 1990.5ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_1_6gf time: 16.4ms (= 1643.8ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.83 (= 16.4ms / 19.9ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_3_2gf time: 25.1ms (= 2505.8ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_3_2gf time: 21.6ms (= 2164.3ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.86 (= 21.6ms / 25.1ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_8gf time: 45.0ms (= 4496.5ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_8gf time: 41.3ms (= 4133.2ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.92 (= 41.3ms / 45.0ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_16gf time: 67.8ms (= 6781.8ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_16gf time: 63.5ms (= 6351.3ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.94 (= 63.5ms / 67.8ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_y_32gf time: 126.5ms (= 12647.5ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_y_32gf time: 107.4ms (= 10735.3ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.85 (= 107.4ms / 126.5ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_400mf time: 12.1ms (= 1210.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_400mf time: 12.8ms (= 1284.8ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
✔️ Relative speed: 1.06 (= 12.8ms / 12.1ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_800mf time: 11.0ms (= 1104.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_800mf time: 9.2ms (= 920.1ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.83 (= 9.2ms / 11.0ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_1_6gf time: 15.6ms (= 1561.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_1_6gf time: 12.2ms (= 1220.7ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.78 (= 12.2ms / 15.6ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_3_2gf time: 23.1ms (= 2310.5ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_3_2gf time: 19.3ms (= 1933.2ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.84 (= 19.3ms / 23.1ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_8gf time: 35.1ms (= 3511.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_8gf time: 31.7ms (= 3168.2ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.90 (= 31.7ms / 35.1ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_16gf time: 62.8ms (= 6279.9ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_16gf time: 57.1ms (= 5712.4ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.91 (= 57.1ms / 62.8ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

OneFlow regnet_x_32gf time: 110.3ms (= 11029.4ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
PyTorch regnet_x_32gf time: 105.3ms (= 10529.6ms / 100, input_shape=[16, 3, 224, 224], backward is disabled, ddp, world size=2)
❌ Relative speed: 0.95 (= 105.3ms / 110.3ms)

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---